the international rice research institute (irri) was

The International Rice Research Institute (IRRI) was established in 1960 by theFord and Rockefeller Foundations with the help and approval of the Governmentof the Philippines. Today IRRI is one of the 16 nonprofit international researchcenters supported by the Consultative Group on International Agricultural Re-search (CGIAR). The CGIAR is sponsored by the Food and Agriculture Organi-zation of the United Nations, the International Bank for Reconstruction and De-velopment (World Bank), the United Nations Development Programme (UNDP),and the United Nations Environment Programme (UNEP). Its membership com-prises donor countries, international and regional organizations, and private foun-dations.

As listed in its most recent Corporate Report, IRRI receives support, throughthe CGIAR, from donors such as UNDP, World Bank, European Union, AsianDevelopment Bank, International Fund for Agricultural Development (IFAD),Rockefeller Foundation, and the international aid agencies of the followinggovernments: Australia, Bangladesh, Belgium, Brazil, Canada, People’s Republicof China, Denmark, France, Germany, India, Islamic Republic of Iran, Japan,Republic of Korea, Mexico, The Netherlands, Norway, Philippines, Spain,Sweden, Switzerland, Thailand, United Kingdom, and United States.

The responsibility for this publication rests with the International RiceResearch Institute.

Copyright International Rice Research Institute 2000

Mailing address: DAPO Box 7777, Metro Manila, PhilippinesPhone: (63-2) 845-0563, 844-3351 to 53Fax: (63-2) 891-1292, 845-0606Email: [email protected]: http://www.cgiar.org.irriRiceweb: http://www.riceweb.orgRiceworld: http://www.riceworld.orgCourier address: Suite 1099, Pacific Bank Building

6776 Ayala Avenue, MakatiMetro Manila, PhilippinesTel. (63-2) 891-1236, 891-1174, 891-1258, 891-1303

Suggested citation:Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors.2000. Characterizing and understanding rainfed environments. Proceedingsof the International Workshop on Characterizing and Understanding RainfedEnvironments, 5-9 Dec. 1999, Bali, Indonesia. Los Baños (Philippines):International Rice Research Institute. 488 p.

Layout and design: Ariel PaelmoFigures and illustrations: Ariel PaelmoCover design: Juan Lazaro IV

ISBN 971-22-0152-X

iii

Contents

FOREWORD vii

ACKNOWLEDGMENTS ix

OverviewCharacterizing rainfed rice environments: an overview of the biophysical aspects 3

V.P. Singh, T.P. Tuong, and S.P. Kam

Characterizing environments for sustainable rice production 33Van Nguu Nguyen

Tools and methodologies for biophysical characterizationEffect of climate, agrohydrology, and management on rainfed rice productionin Central Java, Indonesia: a modeling approach 57

A. Boling, T.P. Tuong, B.A.M. Bouman, M.V.R. Murty, and S.Y. Jatmiko

Perception, understanding, and mapping of soil variabilityin rainfed lowlands of northeast Thailand 75

T. Oberthür and S.P. Kam

Identifying soil suitability for subsoil compaction to improve water-and nutrient-use efficiency in rainfed lowland rice 97

D. Harnpichitvitaya, G. Trébuil, T. Oberthür, G. Pantuwan, I. Craig,T.P. Tuong, L.J. Wade, and D. Suriya-Arunroj

Modeling water availability, crop growth, and yield of rainfed lowland ricegenotypes in northeast Thailand 111

S. Fukai, J. Basnayake, and M. Cooper

Using reference lines to classify multienvironment trials to the target populationof environments, and their potential role in environmental characterization 131

G.C. McLaren and L.J. Wade

Biophysical characterization and mappingBiophysical characterization of rainfed systems in Java and South Sulawesiand implications for research 145

I. Amien and I. Las

iv

Monitoring rainfed and irrigated rice in Southeast Asia usingradar remote sensing 157

R. Verhoeven, H. van Leeuwen, and E. van Valkengoed

Characterizing soil phosphorus and potassium status in lowland andupland rice-cropping regions of Indonesia 169

A. Clough, I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno, and S. Fukai

Planning and managing rice farming through environmental analysis 191K. Borkakati, V.P. Singh, A.N. Singh, R.K. Singh,A.S.R.A.S. Sastri, and S.K. Mohanty

Agroclimatic inventory for environmental characterization of rainfed rice-basedcropping systems of eastern India 215

A.S.R.A.S. Sastri and V.P. Singh

Agrohydrologic and drought risk analyses of rainfed cultivationin northwest Bangladesh 233

A.F.M. Saleh, M.A. Mazid, and S.I. Bhuiyan

Characterizing biotic stressesCharacterizing biotic constraints to production of Cambodianrainfed lowland rice: limitations to statistical techniques 247

G.C. Jahn, Pheng Sophea, Pol Chanthy, and Khiev Bunnarith

Weed communities of gogarancah rice and reflections on management 269H. Pane, E. Sutisna Noor, M. Dizon, and A.M. Mortimer

Socioeconomic characterizationThe role of characterization in ex ante assessment of research programs:a study in the rainfed rice production system 291

P.K. Joshi and Suresh Pal

Constraints to the adoption of modern varieties of rice in Bihar, eastern India 305A. Kumar and A.K. Jha

Rainfed rice, risk, and technology adoption: microeconomic evidencefrom eastern India 323

H.N. Singh, S. Pandey, and R.A. Villano

Using gender analysis in characterizing and understandingfarm-household systems in rainfed lowland rice environments 339

T. Paris, A. Singh, M. Hossain, and J. Luis

Agricultural commercialization and land-use intensification:a microeconomic analysis of uplands of northern Vietnam 371

Nguyen Tri Khiem, S. Pandey, and Nguyen Huu Hong

Economics of intensive rainfed lowland rice-based cropping systemsin northwest Luzon, Philippines 391

M.P. Lucas, S. Pandey, R.A. Villano, D.R. Culannay, and T.F. Marcos

v

Integrating biophysical and socioeconomic characterizationSocioeconomic and biophysical characterization of rainfedversus irrigated rice production in Myanmar 407

Y.T. Garcia, M. Hossain, and A.G. Garcia

Integration of biophysical and socioeconomic constraints in rainfed lowland ricefarm characterization: techniques, issues, and ongoing IRRI research 441

C.M. Edmonds and S.P. Kam

Regional land-use analysis to support agricultural and environmentalpolicy formulation 471

B.A.M. Bouman, R. Roetter, R.A. Schipper, and A.G. Laborte

Foreword vii

Foreword

Rainfed rice areas are associated with a high incidence of poverty, mainly because oflow and unstable yields in a difficult environment where rainfall and water availabil-ity are both seasonal and highly variable, and soil conditions are highly heteroge-neous. These areas have not benefited much from the technological advances of theGreen Revolution, which have targeted the more favorable irrigated environments.Yet even small improvements made in increasing and stabilizing rice yields couldsignificantly alleviate poverty and improve food security among the poorest of thefarming communities in these areas. The challenges to attaining even these smallimprovements are great, however, because of the variable biophysical conditions andsocioeconomic circumstances in these difficult ecosystems.

Over the years, IRRI, in partnership with national agricultural research systemsin several Asian countries, especially the member countries of the Rainfed Lowlandand Upland Rice Research Consortia, has been carrying out studies to understandrainfed rice environments and cropping practices so as to translate research into ap-propriate technological interventions. This research ranges from broad regional-scalecharacterization to detailed farm-level studies. Studies have been and are being car-ried out in different geographical areas by different institutions. Different methodolo-gies have been tested and several cross-cutting issues have emerged.

IRRI, together with the Central Research Institute for Food Crops, Indonesia,organized a thematic workshop on Characterizing and Understanding Rainfed Envi-ronments held 5-9 December 1999 in Bali, Indonesia. The objectives were to reviewprogress on research related to characterizing and understanding rainfed lowland riceenvironments, with emphasis on work done at RLRRC sites, and consider future re-search issues and opportunities for collaboration. Seventy scientists from 15 coun-

viii Foreword

tries gathered to discuss subjects covering these objectives. This book, which waspartially supported by the Australian Centre for International Agricultural Research,contains the papers presented at the workshop. We hope that this will be a usefulsource of information for translating our understanding into appropriate technologi-cal interventions in the highly heterogeneous and variable rainfed environments.

Ronald P. CantrellDirector General

Foreword ix

Acknowledgments

IRRI is most grateful to the Australian Centre for International Agricultural Researchfor its financial contribution to the publication of this book. Many people contributedto the success of the international workshop upon which it is based. The workshopwas co-hosted by the Central Research Institute for Food Crops, Indonesia. The localorganizing committee included Andi Hasanuddin, Sunendar Kartaatmadja, Suprapto,Hasil Sembiring, and Mahyuddin Syam. The overall organizing and technical reviewcommittee included the following IRRI researchers: T.P. Tuong (chair), B.A.M.Bouman, S.P. Kam, S. Pandey, and L. Wade. L.L. Garcia ably handled internationallogistics for the workshop and Hekki provided technical and logistical support to thelocal committee. We thank D. Dawe, C. Edmonds, S. Fukai (Queensland University,Australia), T. George, K.L. Heong, S. Morin, and T. Paris for reviewing papers in thisbook.

Characterizing rainfed rice environments: an overview . . . 1

Overview

2 Singh et al


The literature reveals that rainfed rice environments have been characterizedfor various purposes at differing scales using a range of techniques. Thischapter reviews the biophysical aspects of characterization undertaken inrainfed rice environments, more specifically at sites of the Rainfed LowlandRice Research Consortium. It presents the status of the work done; providesan inventory of techniques applied; discusses scale, variability, and accu-racy; pinpoints gaps in knowledge; and provides future directions for workfrom which these sites can benefit. The chapter brings out commonalities ofcharacteristics across rainfed sites, provides insights into which issues newtechnological developments should focus on when addressing major limita-tions to production enhancement, and identifies opportunities for makingfuture work more efficient and relevant to needs. Examples for each of theseaspects are drawn from case studies in the rainfed rice regions, and specificcases where environmental characterization has made a significant impacton national systems are cited.

Rainfed rice environments are characterized for various purposes. Some of the pre-dominant uses are prioritizing research on a broad scale, extrapolating technology,interpreting multilocation network research, and identifying recommendation zones.Each of these uses requires different degrees of classification subdivisions, from onlya few at broad levels to many at specific levels, using a combination of parametersand available tools and techniques.

Whatever the purpose, characterization in general is supposed to help increasethe productivity of rainfed environments via a better understanding of their proper-ties, particularly in rainfed areas that face a high degree of temporal variability andspatial heterogeneity. It is within this context that rainfed rice research sites, includ-ing the Rainfed Lowland Rice Research Consortium (RLRRC) sites, are character-ized at different scales with corresponding degrees of detail. However, characteriza-tion alone cannot lead to the impact of technology without addressing aspects of

Characterizing rainfed riceenvironments: an overviewof the biophysical aspectsV.P. Singh, T.P. Tuong, and S.P. Kam

4 Singh et al

development, for example, provisions for technology application, such as the avail-ability of inputs, etc.

This chapter focuses on the biophysical aspects of rainfed rice environments. Itreviews some of the broad environmental classifications and systems of rice ecosys-tem analysis, and reviews work done at different sites and the tools and techniquesapplied. The chapter highlights the commonalities across rainfed sites, pinpoints gapsand methodological limitations, and identifies opportunities for making future workmore efficient and relevant to needs.

Broad environmental classifications

Garrity et al (1996) extensively reviewed the systems of environmental classificationand posed a fundamental question: To what extent can rice research rely upon theseveral efforts at broad agroecological classification at the global level, specificallythe climatic classification systems and the FAO agroecological zone studies? Thereare natural advantages in assessing (if this is feasible) the impact of these broad-scaleclassifications, since these efforts are generalized (relatively fewer classes) and widelyknown.

Among the global climatic classifications, those of Koppen (1936), Thornthwaite(1948), Holdridge et al (1971), and Papadakis (1975) have had a wide currency. TheKoppen system recognizes 13 tropical and subtropical climatic zones. TheThornthwaite system identifies 21 classes. These relate locations to general climates,but are broad. For example, Koppen’s class labeled AW is subhumid tropics with oneseason. Papadakis’s work led to a more complex classification system (more than 500classes). Though moving toward comprehensiveness, this sacrificed simplicity.

The FAO agroecological zone system (Kassam et al 1982) includes a broadlydefined climate component based on temperature in the growing period and length ofthe growing period based on soil-water balance. The major objective of the FAOsystem was to assess the suitability of land for different crops (Higgins et al 1987).The major climates can be used independently or combined with growing-period zones.When the climate classification is combined with soils data, it yields a more compre-hensive and complex global agroecological zone (AEZ). In addition to the global andcontinental level studies, the agroecological zone studies have been conducted inseveral countries, for example, Bangladesh (FAO 1988). A world data bank of theAEZ system is maintained at FAO headquarters in Rome. The Technical AdvisoryCommittee (TAC 1990) of the Consultative Group on International Agricultural Re-search (CGIAR) adopted the FAO system of seven basic agroecological units for itsanalysis of CGIAR priorities.

Systems of rice ecosystem analysis

The methods discussed above provide a range of flexibility in data requirement andaggregation and are widely used elsewhere. They have not been widely applied incharacterizing and classifying rice environments, mainly because of the uniqueness


of rice’s environmental situation compared with other major crops. Rice environ-ments are dominated by surface flooding patterns. Therefore, all the rice classifica-tion systems recognize surface hydrology as the dominant delineating variable (Garrity1984). In addition, a meaningful classification of rice environments must proceedindependently of the commonly known global systems.

The agroclimatic classification for rice and rice-based cropping systems thathas been widely adopted (Oldeman 1980, Oldeman and Frere 1982) is based on thelength of the rice-growing season—months in which surface flooding can be main-tained (monthly rainfall of above 200 mm). National agroclimatic maps based on thissystem were derived for several countries, such as the Philippines, Bangladesh, andIndonesia (Oldeman 1980). Huke (1982) compiled maps that uniformly classified allthe countries of South and Southeast Asia in this system. Agro-hydrological factorshave been studied in only some cases: as land qualities in land evaluation for rice-cropping patterns by Tuong et al (1991) at the macro level and by Minh (1995) at themeso-micro level in the Mekong Delta. However, the information on the dynamics ofsurface hydrology is sparse in most of the studies. They also generally do not containinformation on soil type and topography, which are essential land features and arestrongly related to field hydrology as well as to surface flooding patterns to someextent.

The International Terminology of Rice Growing Environments (IRRI 1984)established a standardized scheme of rice ecosystems, which subdivided the com-monly accepted rice environments into a varying number of subecosystems, based onhydrological, temperature, and, in some cases, soil factors (Table 1). Five

Table 1. Terminology for rice-growing environments.

Environment Characteristics

Irrigated Irrigated, with favorable temperatureIrrigated, low-temperature, tropical zoneIrrigated, low-temperature, temperate zone

Rainfed lowland Rainfed shallow, favorableRainfed shallow, drought-proneRainfed shallow, drought- and submergence-proneRainfed shallow, submergence-proneRainfed medium deep, waterlogged

Deepwater Deep waterVery deep water

Upland Favorable upland with long growing season (LF)Favorable upland with short growing season (SF)Unfavorable upland with long growing season (LU)Unfavorable upland with short growing season

(SU)Tidal wetlands Tidal wetlands with perennially fresh water

Tidal wetlands with seasonally or perenniallysaline water

Tidal wetlands with acid sulfate soilsTidal wetlands with peat soils

Source: IRRI (1984).

6 Singh et al

subecosystems were recognized within the rainfed lowland and four within the up-land. Subsequent efforts have attempted to sharpen the classes and provide betterestimates of their overall extent.

Four distinct levels of ecosystems analysis, in terms of geographic scope andmapping scale, were recognized later (IRRI 1992): mega, macro, meso, and micro.Table 2 gives the indicative objectives and activities associated with work at each ofthese levels. Many technical problems still arise in ecosystems analysis and they aredue to the difficulties of working at different mapping scales and transferring infor-mation across scales.

Table 3 presents some of the methods that have been commonly used in evalu-ating the parameters of rice ecosystems and Table 4 lists the parameters that havebeen used for characterization at different scales. These methods have a wide rangeand flexibility in their application and accordingly provide various outputs: from simpledescriptions to semi- or fully quantitative measurements, and computer-simulatedestimates. The techniques employed range from informal and formal interviews andsurveys including reconnaissance, rapid rural appraisals, field visits, and farm/house-hold surveys to simple accounting and bookkeeping, application of procedures forspecific field and laboratory measurements, remotely sensed image and aerial photo-graph interpretation, applications of detailed simulation models of crop growth, andthe use of geographic information systems (GIS). Examples of the tools used arestructured and unstructured questionnaires, field books, specific instruments, and datarecorders. However, the tools and techniques in characterization cannot be separated.They are very closely linked, as a tool is basically an instrument and the applicationof it is a technique. The parameters or factors in the context of characterization arethose properties, obtained from primary and/or secondary sources, that can help de-scribe sufficiently the unit (site/system) in terms of its properties, thereby enhancingthe understanding of the system and serving as a diagnostic criterion to differentiate itinto its subclasses.

While the field surveys provide ground truth as well as other needed ancillaryinformation, the advent of remote sensing and GIS has greatly enhanced character-ization and mapping capabilities. However, these methods are not fixed for any givenlevel of analysis and can also be used at other levels. For example, methods 1 and 2 inTable 3 can also be used at the micro and other levels.

Mega-level analysisWhen IRRI restructured its research programs (IRRI 1989a) to explicitly address therice ecosystems, decision criteria for allocating funds on an ecosystem basis becamemore explicit. The rice area and production in each ecosystem were fundamentalinformation in applying a resource allocation model.

Aggregate data and maps on the rice area by cultural type have been standard-ized for about two decades (Huke 1982, Huke and Huke 1997) on the basis of judi-cious estimates. The accuracy of mega-level data on the amount of rice land in thefour major ecosystems is still uncertain, except in a few countries, because the na-


Tabl

e 2

. In

dica

tive

bas

ic o

bjec

tive

and

act

ivit

ies

in e

cosy

stem

s an

alys

is a

t fo

ur d

iffer

ent

leve

ls o

f an

alys

is.a

Leve

l of

Sca

le o

fU

nit

ofEx

ampl

e of

anal

ysis

Area

(ha

)m

appi

nga

anal

ysis

Obj

ectiv

esAc

tiviti

es p

aram

eter

s/fa

ctor

sin

clud

ed

Meg

a>

1,0

00

,00

01

:50

0,0

00

–R

ice

Iden

tific

atio

n an

d ge

ogra

phic

Bro

ad in

tern

atio

nal-/

natio

nal-

Key

(fe

w);

land

for

m1

:5 m

illio

n e

cosy

stem

sdi

strib

utio

n of

dif

fere

nt e

co-

leve

l ana

lysi

s; c

hara

cter

izin

ghy

drol

ogy,

incl

udin

gsy

stem

s, e

stab

lishi

ng t

heir

and

map

ping

key

eco

logi

cal

flood

ing

and

drou

ght

rela

tive

impo

rtan

ce for

para

met

ers

as m

appi

ngpa

tter

ns;

land

use

;re

sear

ch p

riorit

izat

ion

units

geop

oliti

cal a

nd p

olic

yM

acro

>5

00

,00

0 t

o1

:50

,00

0–

Ric

eU

nder

stan

ding

bro

ad s

patia

lN

atio

nal-

and

regi

onal

-leve

lD

omin

ant;

soi

l and

clim

atic

<1

,00

0,0

00

1:5

00

,00

0

sub

eco-

diff

eren

tiatio

n of

sub

eco-

anal

ysis

; de

taile

d st

udy

ofzo

nes;

land

form

s, fl

oodi

ng,

s

yste

ms

syst

ems.

Tar

get

area

s fo

rec

osys

tem

s, d

iffe

rent

iate

dan

d dr

ough

t cha

ract

eris

tics;

regi

onal

res

earc

h pl

anni

ngus

ing

dom

inan

t fa

ctor

sla

nd u

se; c

ross

-cul

tura

l and

and

prio

ritiz

atio

n; r

esou

rce

as m

appi

ng u

nits

com

mun

ity r

esou

rce

base

,al

loca

-ion

and

broa

d ex

trap

o-ne

eds,

and

pol

icy

latio

n; in

dica

tive

varia

bilit

yw

ithin

eco

syst

ems

Mes

o>

50

,00

0 t

o1

:10

,00

0–

Ric

eS

elec

tion

of r

epre

sent

ativ

eR

egio

nal-

and

zona

l-lev

elS

elec

ted;

soi

l gro

ups,

land

<5

00

,00

01

:50

,00

0

env

ironm

ents

rese

arch

site

s in

dif

fere

ntan

alys

is;

delin

eatin

g ar

eas

type

, dr

ough

t an

d flo

odin

gen

viro

nmen

ts,

extr

apol

atio

nw

ith h

omol

ogou

s co

nditi

ons,

char

acte

ristic

s, r

ainf

all;

and

reco

mm

enda

tion

dom

ains

usin

g se

lect

ed f

acto

rs a

sfa

rmin

g sy

stem

s;m

appi

ng u

nits

; id

entif

icat

ion

com

mun

ity r

esou

rce

base

,of

pot

entia

l res

earc

h si

tes

need

s, a

nd a

ctio

nsM

icro

<5

,00

01

:5,0

00

–R

ice

agro

eco-

Det

erm

inin

g ag

roec

osys

tem

s,M

unic

ipal

- and

vill

age-

leve

lC

omm

on (m

any)

; so

il un

its,

1:2

5,0

00

s

yste

ms

desi

gn o

f on

-farm

exp

eri-

anal

ysis

; de

taile

d st

udy

land

typ

e, h

ydro

logy

(ra

in-

men

ts, se

lect

ion

of r

epre

sen-

cons

ider

ing

all p

reva

lent

fall,

wat

er s

uppl

y so

urce

s,ta

tive

expe

rimen

tal s

ites

for

fact

ors,

tra

nsec

ts,

agro

eco-

surf

ace

wat

er d

epth

,cr

op a

dapt

atio

n, e

xten

sion

syst

em m

aps,

and

sys

tem

sse

epag

e, a

nd p

erco

latio

n);

reco

mm

enda

tions

diag

ram

s; s

elec

tion

of fie

ldcr

oppi

ng p

atte

rns,

yie

lds,

expe

rimen

tal s

ites

varie

ties,

and

pes

tpr

essu

re;

hous

ehol

dre

sour

ce b

ase

and

need

s

a The

se a

re in

dica

tive

figur

es b

ased

on

curr

ent

map

ping

stu

dies

; ad

opte

d fr

om IR

RI (

19

93

).

8 Singh et al

Tabl

e 3

. M

etho

ds c

omm

only

use

d in

eva

luat

ing

para

met

ers

of r

ice

ecos

yste

ms

at d

iffer

ent

leve

ls.a

Met

hod

Des

crip

tion

Leve

l of

Geo

grap

hica

l are

aR

efer

ence

anal

ysis

1.

Para

met

ric●

Inde

x fo

r di

ffer

ent

para

met

ers

ofM

ega

Cou

ntrie

s of

S

outh

and

Gar

rity

and

Agus

tin (1984)

met

hod

clas

sific

atio

n of

upl

and

rice-

grow

ing

Sou

thea

st A

sia

envi

ronm

ent

●Ag

ro-h

ydro

logi

cal p

aram

eter

sM

ega

Mek

ong

Del

ta, Vi

etna

mTu

ong

et a

l (1991)

2.

Rem

ote

sens

ing-

●S

atel

lite

imag

e in

terp

reta

tion

and

Meg

aEn

tire

Cam

bodi

aG

arrit

y an

d B

ruce

(1992)

base

d m

etho

dgr

ound

tru

th●

Sat

ellit

e im

age

inte

rpre

tatio

n an

dM

eso

Mek

ong

Del

taK

am e

t al

(2000)

grou

nd t

ruth

One

dis

tric

t in

Indi

a (4

0 m

ore

IRR

I (1993)

dist

ricts

now

)●

Sat

ellit

e im

age

inte

rpre

tatio

n an

dM

eso

Som

e si

tes

in C

ôte

d’Iv

oire

WAR

DA

(19

93

)gr

ound

tru

th; al

so u

se o

f av

aila

ble

anci

lliar

y in

form

atio

n●

Sat

ellit

e im

age

inte

rpre

tatio

n an

dM

eso

Entir

e C

ôte

d’Iv

oire

WAR

DA

(1993)

grou

nd t

ruth

●Ae

rial p

hoto

grap

h in

terp

reta

tion

and

Mic

roS

ome

rese

arch

site

sIR

RI (

19

89b)

grou

nd t

ruth

for

col

lect

ing

info

rmat

ion

in T

haila

ndon

fie

ld e

xper

imen

tal s

ites

3.

Rap

id r

ural

●H

ouse

hold

and

vill

age

surv

eys,

Mic

roM

ore

than

100 v

illag

esLi

ghtf

oot

et a

l (1989),

Sin

ghap

prai

sal (

RR

A)in

terv

iew

with

ext

ensi

on a

gent

s an

din

eas

tern

Indi

a e

t al

(1993, 1999)

farm

ers,

fie

ld o

bser

vatio

ns a

ndm

easu

rem

ents

, us

e of

sec

onda

ry d

ata

●In

terv

iew

with

ext

ensi

on a

gent

s an

dM

acro

Entir

e C

ôte

d’Iv

oire

WAR

DA

(1993)

farm

ers,

som

e fie

ld o

bser

vatio

ns

a Ado

pted

fro

m S

ingh

and

Sin

gh (

19

95

).


continued on next page

Table 4. Parameters used in some published rainfed rice environmental characterizations and analysisat different scales.

Scale of analysis/characterization Geographical Environmental characteristics/and rice area area parameters studied Referencecovered

Mega South and ● Rainfall in 5-mo growing season Garrity et al (1986)(30 million ha) Southeast ● No. of months R>PE + 20%

Asia ● Soil textural class● Slope class● Soil units (physical and chemical

characteristics and constraintssuch as low CEC, high P fixation,alkalinity and salinity, Zndeficiency; organic, acid sulfate,and shallow soils

Mega Eastern India ● Irrigation IRRI (1993)(26.8 million ● Water depthha) ● Land form

Macro (–) Bangladesh ● Various soil physical and chemical Ahmed et al (1992)properties

● Land use● Rainfall FAO (1988)● Length of growing season● Crop establishment techniques● Water quality

Macro Eastern India ● Rice yield reductions due to major Widawsky and O’Toole(10.6 million ha) biophysical stresses (1990), IRRI (1993)

● Drought● Flooding pattern● Water balance● Length of growing season● Rice yield

Macro Côte d’Ivoire ● Rainfall WARDA (1992)(329,000 ha) ● Toposequence position

● Tillage method● Rice variety, sowing, and inter-

cropping technique● Land tenure● Decision-making by gender● Production objective

Meso Bahraich ● Land form Singh and Pathak(200,000 ha) District, India ● Physiography and slope (1990), IRRI

● Soil texture (1992, 1996)● Soil fertility● Water sources● Irrigation

10 Singh et al


Table 4 continued.


● Groundwater table● Spatial extent under different

environments● Water regime● Floods● Droughts● Rice cultivation practices● Rice yields● Fertilizer and other inputs● Problems and suggestions

Meso Solana, Cagayan, ● Rainfall IRRI (1987, 1990),(176,000 ha) Philippines ● Land form Garrity et al (1992)

● Hydrology● Drainage● Soil properties● Cropping pattern● Rice-farming technology

Meso Faizabad ● Rice area IRRI (1993)(181,000 ha) District, India ● Rainfall pattern

● Irrigation● Land use● Soil constraints● Submergence● Drought● Soil texture and fertility● Groundwater table● Rice area under different

subecosystemsMeso/micro Masodha ● Long-term trend in rice area IRRI (1993)

(21,000 ha) Block, Faizabad ● Rice subecosystems, extent andDistrict, India spatial distribution

● Physiography● Rainfall pattern, flood and

drought occurrence● Soil texture and area coverage● Soil constraints and their

respective area coverage● Drought-affected area● Irrigation source, supply, and

irrigated area



Table 4 continued.


● Available groundwater potentialor development

● Major environmental constraintsto production

Micro (85 ha) Chandpur ● Land types and their area IRRI (1993)Village, Masodha ● Soil texture and fertilityBlock, Faizabad in each land typeDistrict, India ● Groundwater table in each

land type● Flooding● Crops grown in each land type● Farm-related and other enterprises● Farmers’ problem identification and

its influence on rice productivity● Household number and social

group compositionMicro (385 ha) Four villages ● Land type and use pattern IRRI (1991), Singh

in Hazaribagh ● Varieties grown et al (1994)District, India ● Rice yield by land-type varieties

● Social class● Land sharing and division● Credit availability● Water resources and irrigation● Soil characteristics● Pest, disease, and weed pressure● Farming system● System problems

Micro (–) Claveria research ● Land form Magbanua andsite, Northern ● Soils Garrity (1988)Mindanao, ● SlopePhilippines ● Land use

● Farm size and land tenure● Infrastructure● Climate and long-term rainfall

pattern● Grouping of landholding size by

social class● Rice crop area by social class and

landholding size● Rice yields by social class and

landholding size● Fertilizer and other input use by

farm household

12 Singh et al

tional-level statistics on various types of nonirrigated rice lands are generally un-available.

The work of Huke yielded standard maps of the regional allocations of riceland by ecosystem. The maps provided the basis for more comprehensive mega-levelgeographic databases, for focusing characterization on selected micro-regions, andfor classifying them into subecosystems. The upland ecosystem geographic databasewas the initial product (Garrity 1984, Jones and Garrity 1986). This database con-tained data on several agroclimatic and soil parameters for each of the approximately4,000 upland locations on the Huke maps for South and Southeast Asia. The siteswere classified consistently according to a two-factor upland rice environmental clas-sification based on the length of the growing season and inherent soil fertility con-straints, and also a three-factor system that included an estimate of seasonal moisturesufficiency. The two-factor classification conforms with the four broad uplandsubecosystems specified in the International Terminology of Rice Environments (IRRI1984). The three-factor classification recognizes 12 major classes and, at a more de-

Table 4 continued.


Micro (–) Rainfed lowland ● Land form IRRI (1989b)rice research ● Physiographysite, Khukhar ● SlopeVillage, Thailand ● Land use

● Soil fertility and othercharacteristics

● Detailed field hydrology● Groundwater table● Rice yields in different soil and

land combinationsMega South and ● Climate-rainfall, temperature, and Garrity (1984)

(11.6 million Southeast Asia growing season length ha) ● Slope, soil texture, soil groups,

and inherent fertility statusMacro Eastern India ● Irrigation extent IRRI (1993)

(26.8 million ● Land formha) ● Fertility-related constraints

Macro Côte d’Ivoire ● Rainfall WARDA (1992)(329,000 ha) ● Toposequence position

● Tillage method● Rice variety, sowing and inter-

cropping technique● Land tenure● Decision-making by gender● Production objective


tailed level, that is, by using three categories of growing-season length, six categoriesof soil fertility constraints, and four categories of water balances, a total of 72 classes.

The rainfed lowland rice ecosystem database (Garrity et al 1986) was compiledusing a similar methodology. Approximately 6,300 rainfed lowland sites were classi-fied in a three-factor environmental classification that included growing-season length,water balance, and soil constraints as delimiters (Garrity 1984). The other mega-levelcharacterization study is of eastern India (IRRI 1993), with a similar basis of assess-ment.

The Asian rice-land soil constraints database covers all rice land in South andSoutheast Asia, including irrigated and deepwater area (IRRI 1987). This databaseincludes data from the FAO Soil Maps of the World (FAO 1977, 1979), with soilconstraints interpreted according to the fertility capability classification (Buol andCuoto 1981).

These databases were intended to be useful in research prioritization. Theirimpact has been significant in terms of a major shift in upland breeding and agro-nomic research in the early 1980s, from recent volcanic soil to acid upland soils, andfrom flat land to sloping land.

The rainfed lowland database had a lesser impact initially most likely becauseit did not include data on surface water depth regime, particularly the frequency andduration of crop submergence. Although the database included length of the growingseason and a crude water balance classification, sufficient data did not exist to clas-sify the surface water accumulation dynamics at the micro-region level. In its ab-sence, there was no way to definitely classify and map the rainfed lowlands into thefive subecosystems. However, with the development of remote-sensing and GIS-basedmethodology (Singh 1987, 1988), the assessment of flood as well as drought becameeasily possible. Further development of this methodology (Singh and Singh 1996),which uses the crop vegetation index as a comprehensive reflection of prevailingconditions, including hydrological ones, in the crop, allowed taking full account ofseasonally variable environmental conditions and, thus, the reliable classification andmapping of rainfed rice lowlands into various subecosystems. This methodology,coupled with the rainfed lowland database, is now being used extensively throughouteastern India to analyze and map about 21 million ha of rainfed rice into subecosystems,and is presented in the next sections.

Macro-level analysisThe generalized nature and small scale of mega-level databases strongly limit theirapplication beyond regional issues. Many nations are interested in researchprioritization and extrapolation that typically require much more detailed informa-tion, at least at the macro level (national/state level for large states). In several coun-tries, some remarkably comprehensive and useful data sets have been developed (FAO1988, Widawsky and O’Toole 1990, IRRI 1993, WARDA 1992). The challenge is tomake better use of these data. An excellent example is the case of Bangladesh (Ahmedet al 1992), where extensive soil and land-use data sets and maps were developed bythe Soil Resources and Development Institute in collaboration with FAO (FAO 1988).

14 Singh et al

This included standard countrywide data on surface flooding regime, data which arerare in most countries.

The entire rainfed rice area in eastern India, covering the states of Assam, Bihar,and West Bengal and eastern parts of Uttar Pradesh and Madhya Pradesh, has beendelineated in different rice ecosystems using the information on rice cultural typescoupled with growing-season length, a field water balance, and soil quality (Singhand Sastri 1998, IRRI 1993). An agroclimatic analysis of this region includes data-bases on major climatic variables, a full account of the climatic water balance, and adetailed agronomic interpretation of planting schedules, varieties to be used, and othercrop management practices according to moisture availability. Analysis of rice areasin different eco-classes has provided more reliable estimates of their extent and as-sisted the government in setting priorities.

An agroclimatic atlas of eastern India depicting this analysis and detailing theclimatic water balance for 45 stations in the region has been prepared (Singh et al1999) as a reference for research and development agencies. Other such efforts at theregional level, primarily looking into hydrological aspects, are those of Tuong et al(1991) and Kam et al (2000).

Meso-level analysisA major positive trend in many countries is the regionalization of research acrosssubnational border linkages, for example, across districts/states in India. Nationaland subnational governmental efforts have enabled the development of institutionsthat can identify the unique problems and research priorities of the specific areaswhere they are located. These institutions seek methods to establish priorities that aresuited to smaller geographic areas and larger mapping scales (Garrity et al 1996). Theprocess is more suited to direct feedback from extension personnel and on-farm adop-tive research.

Meso-level analysis is typically associated with a cultivated area of about 100,000hectares, using a mapping scale of 1:25,000 to 1:100,000. Analysis at this scale canefficiently identify and delineate rice ecosystems and subecosystems in terms of sur-face hydrology, land form, and soil classes. Associated with each rice ecosystem arethe flood and drought frequencies and duration, prevalent cropping patterns, and cropmanagement practices.

An example of a useful meso-level analysis was that conducted for BaharaichDistrict, Uttar Pradesh, India, to identify the problems causing low rice yields and toprepare the priority research agenda at a district level in eastern India (Singh andPathak 1990). The analysis determined the ecological variability of the district interms of hydrology (rainfall pattern, water table depth, irrigation sources, and drain-age), land form and slope, length of the growing season, frequency and duration offlood and drought, and major insect, disease, and weed pressures.

After characterization, the factors were combined to identify and delineate thearea into homologous zones. Cropping pattern, varieties used, crop management prac-tices and input use, and socioeconomic conditions were superimposed separately on


each of the rice ecosystem maps. The rice ecosystems were then prioritized for re-search on the basis of extent of area, number of affected households, and potentialpossibilities of research success. Similar characterization was done in representativedistricts in each of the eastern India states (IRRI 1992, 1993, 1996) and was used todevelop a comprehensive rice research plan for eastern India.

Faced with the hydrological complexities in the rainfed situations, Singh andSingh (1996) developed a remote-sensing and GIS-based methodology that couldreliably delineate different rice ecosystems and their subecosystems, particularly bytheir hydrology—drought-prone, submergence-prone, etc. Using this methodology,40 of the 93 rainfed rice-growing districts in eastern India have been characterizedand mapped into the principal rainfed lowland subecosystems (unpublished) by teamsof state agricultural universities, Indian Council for Agricultural Research centers,state departments of agriculture, remote-sensing application centers, nongovernmentorganizations, and groups of local farmers. This analysis is already being used exten-sively in regional research planning and to develop and delineate application do-mains of the promising technologies in eastern India (Singh and Sastri 1998). Ex-amples of these efforts are Sastri and Singh (this volume) and Borkakati et al (thisvolume).

IRRI and the Department of Agriculture Regional Office for the Cagayan Val-ley, Philippines (Region II), developed a meso-level classification of the valley’s com-plex mosaic of rainfed rice lands (IRRI 1987, 1990, Garrity et al 1992). They ex-plored the utility of a computerized geographic database correlated with village-levelmaps of rainfed rice land types. The information was packaged as a field manual forextension personnel. Six rainfed rice subecosystems were recognized on a hydrologi-cal basis. They were explicitly correlated with a range of associated information tospecify their identification and the technology associated with them. The data on ricearea and the yield constraints associated with each rainfed rice land type have facili-tated regional rice research efforts, particularly the relative emphasis given to appliedand adaptive research among land types (Garrity et al 1992, IRRI 1991).

Micro-level analysisAgroecosystem analysis has become very popular in micro-level prioritization(Conway 1986, KEPAS 1985). Micro-level analysis studies have been done in sev-eral cases (IRRI 1989b, 1991, 1993, Singh et al 1994, Magbanua and Garrity 1988).This level of analysis has been used extensively in the rainfed regions of easternIndia, covering the states of Uttar Pradesh, Madhya Pradesh, Bihar, Assam, WestBengal, and Orissa, to set research priorities within and among dominant rice-farm-ing systems. Various techniques of agroecosystem analysis such as site descriptions,problem diagnosis, farming systems analysis, and rapid rural appraisal methodologywere extensively employed (Lightfoot et al 1990). The methodology involved a two-tier training program for researchers on the methodology for setting research priori-ties by agroecosystem analysis with farmer participation. The analysis was carriedout by 15 research centers in the region covering upland, rainfed lowland, and

16 Singh et al

deepwater rice ecosystems. The research diagnosis and prioritization at this level wereconducted by multidisciplinary teams in the respective centers, with regular involve-ment and interaction from groups of farmers (IRRI 1989b).

The micro-level agroecosystem analyses (100 locations in eastern India) in-cluded detailed characterization and classification of the static and variable factorsthat differentiate agroecosystems by soils, hydrology, farming practices, and socio-economic conditions (IRRI 1989b, 1993, Singh et al 1993). The sites were mappedon the scale of 1:2,000–1:5,000. At all sites, the static factors studied were land types,landholding size, source of water supply, and soil properties. The dynamic factorswere land use; field water depth; rainfall and cropping patterns; crop yields; varietiesand management practices; insects, diseases, and weeds; production costs and re-turns; labor supply pattern; prices, assets, and income distribution; and demographyby social class.

The geographic area was zoned into agroecosystems and the problems and op-portunities elucidated in each major agroecosystem. Among the differentagroecosystems, the highest priority was given to the one with the largest extent. Theresearch problems were then prioritized on the basis of their physical extent withinagroecosystems (coverage), number of affected households, complexity of the prob-lem, severity of the problem (crop loss estimates), frequency of problem occurrence,and the importance of the affected enterprise in the farming system (IRRI 1989b).This type of analysis has greatly facilitated on-farm research for the development ofproblem-solving technologies, the selection of representative research sites, and theactual adoption/impact of technologies in the region (Singh and Singh 2000). Thesemethodologies are also increasingly being used in Thailand, Bangladesh, Indonesia,and Bhutan, as is reflected in the requests for training on them by the respectivecountries and their research programs developed through these methods.

The characterization studies carried out at the consortium (RLRRC) sites aremostly diagnostic in nature and in detail (micro, field level, Table 5). Consortiumsites serve as the “hot spots” for research on certain thematic issues, such as low-Psoils, and their reports show that they lack the characterization of other existing con-ditions. This is not site characterization but the characterization of a certain theme,such as submergence or drought, at the consortium site. The characterization of onlycertain themes without consideration of other existing conditions at the research sitelimits the utility of research results because the technological output cannot be ex-trapolated without a complete description of the research as well as the target site.This underutility points out the need for a complete characterization of a site that willinclude certain major themes along with other prevailing conditions. Also, a higherlevel (e.g., regional) analysis is carried out only at some consortium sites, such asFaizabad and Cuttack (Table 5). However, higher level analysis has been scheduledto be carried out at other sites, such as Jorhat, Assam, India; the Mekong Delta andRed River Basin, Vietnam; and northeastern Thailand, which are affiliated with theRLRRC key sites. It is essential to foster functional linkages between key and affili-ated sites to ensure effective interchange and to enhance the utility of research.


cont

inue

d on

nex

t pa

ge

Tabl

e 5

. C

hara

cter

izat

ion

at s

ome

cons

orti

a si

tes.

Mai

nM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nS

cale

pre

sent

lyde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

appl

ied

need

s/ga

ps

Ban

glad

esh

●R

ajsh

ahi

Dro

ught

Det

erm

inin

g w

ater

Fiel

d/cr

opM

oist

ure

avai

labi

lity

Cro

ppin

g sy

stem

sav

aila

bilit

yto

pose

quen

ceas

sess

men

tte

chno

logy

tech

niqu

esde

velo

ped

and

adop

ted

●N

arsi

ndhi

,S

easo

nal

Floo

ding

pat

tern

s,D

rain

age

basi

nC

ropp

ing

syst

ems

Kis

hore

Gan

jflo

odin

gde

pth,

and

tech

nolo

gy(f

lood

-pro

nedu

ratio

nde

velo

ped

and

syst

ems)

adop

ted

Indi

a●

Cut

tack

Sub

mer

genc

e/D

eter

min

ing

moi

stur

eW

ater

shed

/Fl

ood

qual

ity a

nddr

ough

tav

aila

bilit

y pe

riod

bloc

ksc

reen

ing

and

flood

wat

erdr

aina

gete

chni

ques

qual

ity m

onito

ring

basi

n●

Faiz

abad

Dro

ught

and

Det

erm

inin

g m

oist

ure

Wat

ersh

ed/

Wat

er b

alan

ceO

n-fa

rmPr

omot

ion

ofsu

bmer

genc

e;av

aila

bilit

y pe

riod

bloc

k dr

ain-

mon

itorin

g/te

chno

logi

espr

omis

ing

crop

ping

and

flood

wat

erag

e ba

sin

scre

enin

gde

velo

ped

tech

nolo

gies

syst

em;

qual

ity m

onito

ring

tech

niqu

esan

d ad

opte

dre

spon

se o

fric

e ge

noty

pes

●H

azar

ibag

hS

hallo

w,

Anal

ysis

of so

ilFi

eld/

cros

sP-

man

agem

ent

Ric

e va

rietie

sR

egio

nal a

naly

sis

P-de

ficie

nt,

chem

ical

topo

-al

tern

ativ

es,

deve

lope

dof

spa

tial

and

acid

icpr

oper

ties,

sequ

ence

deve

lop

and

adop

ted

dist

ribut

ion

soil;

dro

ught

drou

ght,

and

P-ef

ficie

ntby

far

mer

sso

il ph

ysic

al a

ndan

d de

ep-

hydr

aulic

pro

pert

ies

root

ed g

enot

ypes

18 Singh et al

cont

inue

d on

nex

t pa

ge

Tabl

e 5 c

onti

nued

.

Mai

nM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nS

cale

pre

sent

lyde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

appl

ied

need

s/ga

ps

●Jo

rhat

Sub

mer

genc

eFl

oodi

ng p

atte

rn,

Wat

ersh

ed/

Floo

d qu

ality

Cro

ppin

g sy

stem

sPr

omot

ion

ofde

pth,

and

bloc

kan

d sc

reen

ing

tech

niqu

epr

omis

ing

dura

tion

drai

nage

tech

niqu

esde

velo

ped

and

tech

nolo

gies

basi

nad

opte

dIn

dia

●R

aipu

rD

roug

htD

eter

min

ing

moi

stur

eFi

eld/

farm

er/

Wat

er b

alan

ceO

n-fa

rmPr

omot

ion

ofav

aila

bilit

y pe

riod

wat

ersh

edm

onito

ring

and

tech

nolo

gies

prom

isin

gcr

op p

lann

ing

deve

lope

dte

chno

logi

esan

d ad

opte

dIn

done

sia

●S

itiun

gS

ubso

il ac

idity

Cha

ract

eriz

ing

Fiel

dS

tudy

of le

achi

ng o

fEc

onom

ical

Gen

eral

soi

lin

upl

ands

leac

hing

of

surf

ace-

appl

ied

subs

oil a

cidi

typh

ysic

al a

ndca

tions

and

ani

ons;

lime

and

limin

gam

elio

ratio

nch

emic

alw

ater

mov

emen

t,ef

fect

on

subs

oil

for

upla

nds

mea

sure

men

tsso

il ac

idity

root

pro

file

Low

soi

l-P s

uppl

yC

hara

cter

izin

g lo

ng-

Fiel

dLo

ng-te

rmEc

onom

ical

PM

onito

ring

of P

in a

cid

upla

ndte

rm s

oil-P

phos

phor

usm

anag

emen

tdy

nam

ics

need

sso

ilsdy

nam

ics;

soi

lex

perim

ent;

stu

dyst

rate

gies

for

to b

e co

ntin

ued

toph

ysic

al a

ndof

cum

ulat

ive

upla

nds

bett

er u

nder

stan

dch

emic

alef

fect

s of

PP

build

up/d

eclin

ech

arac

teriz

atio

nap

plic

atio

nsef

fect

sEx

tent

of

PD

istr

ict/

On-

farm

tes

ting

ofS

ite-s

peci

ficN

ewly

pro

pose

dlim

itatio

n an

dre

gion

al/

phos

phor

usad

apta

tion

ofra

nge

of p

oten

tial

land

slo

pede

cisi

onP

man

agem

ent

crop

res

pons

es t

ocl

ass

supp

ort

syst

emre

com

men

da-

P ap

plic

atio

n in

(PD

SS

); G

IStio

ns a

nd p

olic

yac

id u

plan

ds o

fch

arac

teriz

atio

nus

ing

PDS

SIn

done

sia

of so

il-P

stat

usan

d cr

opre

spon

ses

Soi

l phy

sica

l and

Fiel

dS

oil s

ampl

e an

dhy

drau

licm

easu

rem

ent

prop

ertie

s


Tabl

e 5 c

onti

nued

.

Mai

nM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nS

cale

pre

sent

lyde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

appl

ied

need

s/ga

ps

Indo

nesi

a●

Jake

nan

Hig

h po

pula

tion

Fiel

dD

ry-s

eede

d ric

eC

rop

Low

and

un-

Mul

tisca

le w

ater

Fiel

dM

onito

ring

wat

erN

eed

for

shor

t-R

egio

nal d

ata

onst

able

yie

ldba

lanc

e, t

akin

gta

ble

dept

hs;

dura

tion

wat

er t

able

and

of t

he s

econ

din

to a

ccou

nt w

ater

wat

er b

alan

ceva

riety

wat

er b

alan

ce:

rain

fed

rice

tabl

e de

pth

and

mon

itorin

g an

dst

ress

edth

eir

dyna

mic

sdi

ffer

ent

sim

ulat

ion

and

spat

ial

topo

sequ

ence

dist

ribut

ions

Cha

ract

eriz

ing

Fiel

dW

ater

bal

ance

and

On-

farm

Soi

l phy

sica

l and

land

scap

e an

dec

onom

ics

ofte

chno

logi

eshy

drau

lic d

ata,

soil

prop

ertie

s fo

ron

-farm

res

ervo

irsde

velo

ped

mic

ro-c

atch

men

tth

e ad

optio

n of

and

adap

ted

char

acte

ristic

s,on

-farm

res

ervo

irsby

far

mer

sre

gion

alap

plic

abili

tyN

utrie

nt d

ynam

ics

Long

-term

Kno

wle

dge

onR

egio

nal d

ata

onan

d its

spa

tial

expe

rimen

ts;

nutr

ient

topo

sequ

ence

dist

ribut

ion

G ×

E a

nddy

nam

ics

and

nutr

ient

nutr

ient

× w

ater

enha

nced

;av

aila

bilit

y an

dex

perim

ent

farm

ers’

dyna

mic

snu

trie

ntm

anag

emen

tim

prov

edH

igh

inte

nsity

Varie

ty s

cree

ning

Fiel

dN

ew v

arie

ties

and

low

inpu

tan

d ad

optio

nre

leas

edle

ad t

o nu

trie

ntm

inin

g,es

peci

ally

K a

nd P

cont

inue

d on

nex

t pa

ge

20 Singh et al

Tabl

e 5 c

onti

nued

.

Mai

nM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nS

cale

pre

sent

lyde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

appl

ied

need

s/ga

ps

Prof

use

wee

dW

eed

dyna

mic

sFi

eld

Wee

d su

rvey

Wee

d dy

nam

ics,

infe

stat

ion;

in d

iffe

rent

topo

sequ

ence

,hi

gh p

est

topo

sequ

ence

popu

latio

n in

infe

stat

ion

and

crop

ping

diff

eren

tpa

tter

nsag

rohy

drol

o-gi

cal d

ynam

ics

Phili

ppin

es●

Bat

acS

oil n

utrie

ntN

utrie

nt d

ynam

ics

Fiel

dN

utrie

nt b

alan

ceB

ette

r nu

trie

ntW

ide-

scal

esu

stai

nabi

lity

stud

ym

anag

emen

tad

apta

tion

in h

ighl

yre

com

men

-in

tens

ified

datio

nsan

d di

vers

ified

crop

ping

syst

ems;

grou

ndw

ater

Gro

undw

ater

qua

lity

Fiel

dG

roun

dwat

erAw

aren

ess

ofR

egio

nal

sust

aina

bilit

yan

d qu

antit

ym

onito

ring

and

wat

er q

ualit

yid

entif

icat

ion

mod

elin

gpr

oble

m;

of a

reas

redu

ce a

mou

ntpr

one

toof

fer

tiliz

erco

ntam

inat

ion

appl

ied

Econ

omic

Farm

er in

com

e an

dFa

rmFa

rm s

urve

y of

sust

aina

bilit

ym

arke

t st

ruct

ure

crop

ping

seq

uenc

ean

d pr

oduc

tivity

●C

lave

ria/

Acid

ic, P-

defic

ient

Nut

rient

dyn

amic

sFi

eld

Scr

eeni

ngK

now

ledg

e of

Extr

apol

atio

nC

avin

tiso

ilte

chni

ques

nutr

ient

to o

ther

are

asdy

nam

ics

enha

nced

and

scre

enin

gte

chni

ques

impr

oved

cont

inue

d on

nex

t pa

ge


Tabl

e 5 c

onti

nued

.

Mai

nS

cale

pre

sent

lyM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nap

plie

dde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

need

s/ga

ps

Phili

ppin

es●

Iloilo

Sho

rt r

ainy

Cha

ract

eriz

e so

ilFi

eld/

farm

/O

n-fa

rm a

ndD

oubl

e ric

ese

ason

,qu

ality

ana

lysi

spr

ovin

cepr

ovin

cial

-leve

lcr

oppi

ng u

nder

drou

ght,

poo

rof

dro

ught

map

ping

deve

lopm

ent

soil

qual

ity,

and

adap

ted

low

cro

ppin

gby

far

mer

sin

tens

ity●

Tarla

cLo

w a

nd u

nsta

ble

Mul

tisca

le w

ater

Fiel

d an

dM

onito

ring

wat

erAg

rohy

drol

ogy

ofR

egio

nal d

ata

yiel

d du

e to

bala

nce,

tak

ing

acro

ssta

ble

dept

hs;

dry-

seed

edon

wat

er t

able

drou

ght

into

acc

ount

topo

-w

ater

bal

ance

rice

syst

eman

d w

ater

wat

er t

able

dep

thse

quen

cem

onito

ring

and

iden

tifie

dba

lanc

e:an

d di

ffer

ent

sim

ulat

ion;

G ×

Eth

eir

dyna

mic

sto

pose

quen

ceex

perim

ents

and

spat

ial

dist

ribut

ions

Low

nut

rient

Nut

rient

dyn

amic

sFi

eld

Nut

rient

by

wat

erK

now

ledg

eR

egio

nal d

ata

onan

d its

spa

tial

expe

rimen

tge

nera

ted

nutr

ient

dist

ribut

ion

avai

labi

lity

and

dyna

mic

sTh

aila

nd●

Ubo

nD

roug

htW

ater

bal

ance

,Fi

eld

Wat

er b

alan

ceS

uita

ble

Anal

ysis

of ta

rget

moi

stur

ege

noty

pes

area

sav

aila

bilit

yde

velo

ped

Viet

nam

●U

plan

d R

edD

roug

htD

eter

min

ing

wat

erFi

eld

Anal

ysis

of

Pote

ntia

lR

egio

nal a

naly

sis

Riv

er B

asin

avai

labi

lity

seco

ndar

yge

rmpl

asm

of t

empo

ral

(Vie

tnam

)pe

riods

clim

atic

dat

a;id

entif

ied

and

spat

ial

in s

itu w

ater

dist

ribut

ion

bala

nce

stud

y;of

dro

ught

varie

ty t

estin

g

cont

inue

d on

nex

t pa

ge

22 Singh et al

Tabl

e 5 c

onti

nued

.

Mai

nS

cale

pre

sent

lyM

etho

dolo

gyC

ount

ryS

iteM

ain

issu

esch

arac

teriz

atio

nap

plie

dde

velo

ped/

Impa

ctU

nfill

edne

eds

appl

ied

need

s/ga

ps

Nut

ritio

n (lo

w P

)Q

uant

ifyin

g so

ilFi

eld

Nut

rient

Bet

ter

nutr

ient

Reg

iona

l soi

l map

chem

ical

man

agem

ent

man

agem

ent

that

can

be

prop

ertie

sex

perim

ents

usef

ul f

rom

agro

nom

icpo

int

of v

iew

Hig

h po

pula

tion

Land

-use

and

land

-Fa

rm,

Sur

vey;

eco

nom

etric

Polic

yM

odel

s fo

r lo

ng-

pres

sure

tenu

re p

olic

ies

hous

ehol

dm

odel

ing

reco

mm

enda

-te

rm p

olic

ytio

ns m

ade

anal

ysis

Food

sec

urity

for

Fact

ors

affe

ctin

gFa

rm,

Sur

vey;

eco

nom

etric

Polic

yM

odel

s fo

r lo

ng-

upla

ndfo

od s

uppl

y an

dho

useh

old

mod

elin

gre

com

men

da-

term

pol

icy

dem

and

tions

mad

ean

alys

is


General discussion

The complexities in biophysical characterization and classification of rainfed envi-ronments stem from the existence of different ecosystems and/or their classes in acontinuum, unclearly defined boundaries between them, overlapping or transitionalzones between the subsystems, and the inability of the methodologies and parametersused to discriminate among them.

In many places, rice ecosystems and/or their classes co-exist, even at short physi-cal distances, sometimes even within a farm. Overlaps or transitional zones also oc-cur between the subsystems, which shift from one class to another from season toseason. Recognizing these facts can be helpful in dealing with the characterizationmethodology and technologies for such areas. This also brings out the limitation ofzoning with fixed boundaries, particularly at more detailed resolution and for tech-nology targeting.

The characterization done either by using diverse parameters, land type in somecases and other parameters in other cases, such as upland and lowland ecosystems byland type and deepwater ecosystems by water depth, or by using related parameters,but exclusive of each other, has posed difficulties in the synthesis of interpretation inan integrated manner. This is particularly relevant where the effect of prevailing envi-ronmental conditions is to be related to crop performance and ways and means are tobe developed for improving it. Otherwise, the results of characterization would re-flect an “ex post facto”) scenario wherein no mid-term corrective measures could bedevised or applied. Therefore, we suggest that different systems be characterized byusing similar and related parameters. This should include a reflection of crop condi-tions and synthesis of the interpretation in a comprehensive manner.

In which eco-category any given geographic location would be under whenenvironmental conditions are variable often can’t be predicted reliably by using con-ventional analytical approaches. Because of this weakness, the remote-sensing andGIS-based methodology (Singh and Singh 1996), which uses the crop vegetationindex (a comprehensive reflection of prevailing conditions and their effect on thecrop) as a major criterion, shows good promise and may be pursued further for thereliable delineation of an area in specific ecologies under variable conditions fromseason to season. The availability of optical remote-sensing data may, however, be alimitation in tropical areas during the monsoon season. Radar-based sensory data arehelpful in overcoming this limitation and they are increasingly becoming availablefor such areas.

The above cases show how the methodologies evolved over time and how cur-rent work, which is site-specific, has benefited from past work. But, in doing so,issues emerge related to new perspectives, new demands, and new tools; scale, vari-ability, and accuracy; parameters that are often overlooked at different levels; andoutsourcing skills, data sharing, and partnership. These issues are related to the fun-damentals of characterization:

1. It is a sequential and in some cases multientity (institution/agency) activitywherein the participating agencies may have different mandates or different

24 Singh et al

levels of involvement. For example, United Nations agencies such as FAOmay have a global mandate, whereas the CGIAR centers may have a com-modity-specific mandate or an ecoregional mandate, and the country/state/provincial organizations may have a national or subnational mandate.

2. It can be an activity that is done to fulfill an objective, such as when technol-ogy development is an objective and characterization is a means for achiev-ing that objective. Impact, of course, is an outcome of the application oftechnology for which the agroecological requirements of the technologyshould be known and conditions suitable for its application identified.

The data requirement, techniques and tools used, and responsibilities for doingactivities will mainly depend on the above two considerations. Other issues are dis-cussed below.

Shifts in emphasisAs has been noted in Tables 4 and 5, the perspectives of characterization have shifted.One such shift is from broad or general uses of characterization to detailed and spe-cific uses through sharper tools and techniques, such as from developing broadlyadapted plants to developing plants for target environments. The other shift has beenin the ways characterizations are used. Characterization can be used for technologyextrapolation and recommendation domains and as a tool for diagnostics in technol-ogy development. It is in this context when scaling up or scaling down the character-ization becomes important, that is, interpreting and interrelating the information acrossscales. There is a need to bridge these two perspectives. The operative support scale,particularly for technology targeting and development, would be the scale at which(1) the technology will have an impact and (2) common interest groups (users, localadministrative units) can effectively take actions.

Data sourcing and sharingThis is a most sensitive and important issue, particularly if the demand is high andurgent. It is also voluminous and involves sensitive information, such as topo sheets,national borders involving national/geopolitical sensitivity, etc. Many national agri-cultural research systems (NARS) are reluctant to give their data to outside agencies;however, they are willing to share the processed information if data processing isdone in their own places. The issue of data sharing is not confined to NARS versusinternational agencies. In many countries, institutional barriers constrain data sharingeven within the country, but it is easier to resolve this among agencies within a coun-try by involving higher authorities. In both cases, a discussion among the partnersfrom the beginning, clearly outlining the benefits to all involved, and devising anagreed-upon procedure to follow may be quite useful in overcoming this problem.

Scale, accuracy, and integration● Inadequacies in techniques and structure of the classification systems. In

the literature, there appears to be no specified classification system or crite-ria for different levels of analysis, except that of IRRI (1984, 1993). There-


fore, any site analyses, even at a particular level, may be carried out usingdifferent sets of parameters on account of their relative importance and re-searchers’ access to the databases. Although this provides much flexibilityto researchers, it runs the risk of resulting in differential classification, whichcould contribute to inappropriate research and development planning for theecology represented by the site. This inadequacy points out the need to fol-low a well-defined set of classification criteria that operate at well-definedspatial scales. We suggest that such criteria be developed through a “work-shop-mode” agreement. A framework developed, from among others, byGarrity et al (1986, 1996), IRRI (1984, 1993), and Jones and Garrity (1986)is provided for consideration in Figure 1 and Table 2. Such a classification,which uses multiple criteria in a hierarchical manner, reflecting the system’sproperties, has been found quite useful and is suggested for use in the future(Fig. 1).

● Lacking classification criteria and parameters for each level. In spite of theuniqueness of rice’s environmental situation (surface hydrology), which de-pends on land characteristics, the land properties (parameters) used in rice

Fig. 1. Relative features of various levels of agrosystems analysis.

Factors: Biophysical (soil, pedon, water and micro- climate)

Biophysical (soil units, water and micro- climateEconomic (household resource base)

Biophysical (soil units, land, hydrology, climate)Socioeconomic (household resource base and needs)

Biophysical (soil groups, land, hydrology, climate, vegetation)Socioeconomic (community resource base, needs, and actions)

Biophysical (soil and climate zones, land, hydrology, vegetation)Socioeconomic (cross-cultural, community resource base, needs, and policy)

Biophysical andsocioeconomic (same as for agroclimatic zone level and geopolitical)

Levels: Soil level (pedon, pot, experimental plot)

Farm level Watershed level Agroclimate zone level

Agroecological zone/ ecoregional level

Systems: Soil-plantsystem

Field-cropsystem

Agro-systems

(farm-householdand otherrelatedenterprises)

Agroclimatic-socioculturalsystems

(multiwater-sheds, agro,forest, socio-cultural, etc.)

Agroeco-logical-sociopoliticalsystems

(geo, agro,nonagro,policy, etc.)

Scale:units:

Micro<1–20 m2

Micro0.1–0.5 ha

Micro0.5–5.0 ha

Micro<5,000 ha

Meso>50,000 ha

Macro>50,000 ha

Mega>1,000,000 ha

(Micro Meso)(5,000–<50,000 ha)

(Meso Macro)(>50,000–<500,000 ha)

Focus: Temporal Temporal Spatial andtemporal

Spatial andtemporal

Spatial andtemporal

Spatial andtemporal

(Macro Mega)(>500,000–<1,000,000 ha)

Agroecosystems

(agro, agro-forest, agro-aqua,nonfarm andoff-farmenterprises)

Field level

26 Singh et al

ecosystem classifications are diverse. Surface hydrology and land character-istics have often been used exclusive of each other.

The lack of specific criteria available in the classification systems hasalso hindered the reliable differentiation of the systems. For example, byusing only drought occurrence as a criterion, Ubon (Thailand), Tarlac (Phil-ippines), and Raipur (Madhya Pradesh, India) are classified only as drought-prone rainfed lowlands, despite the vast difference in their biophysical char-acteristics (soil type, rainfall, amount and pattern of water retention, etc.).Such a classification reflects only the hydrological effect and assumes thatits generic causes are the same across sites, or does not take into account thecauses at all. Because of this, all the areas classified in a particular ecologyappear to be homogeneous in all respects across locations, which is far fromreality. This also indicates (proposes) the application of the same remedialmeasures to overcome the stress across locations.

This issue mostly relates to aspects of scale and the spatial heteroge-neity that exists at every level. The issue here really is of the level of under-standing that is possible and needed. A better understanding is always morehelpful and can be achieved only through detailed analysis. Therefore, theanalysis of the spatial pattern of both the nature of occurrence of stress (e.g.,drought in early, mid, and late season or at different crop growth stages) andseverity of stress (e.g., degree of drought and its temporal feature and effecton the crop) should be included in the characterization, which can providemore meaningful interpretations, especially on management aspects, thanonly classification of the site in an ecological class.

● Unrelated criteria/parameters used in characterization at different levels.Even within a location/site, the parameters used at different levels are notonly different but also generically diverse. In such cases, the analysis at dif-ferent levels cannot be integrated. Table 4 shows an example of parameterdiversity used at different levels. The information in this table also showsthat, in certain types of analysis, there is no direct relationship of any param-eter used in one level to the parameters used in the next level. Therefore, theuse of related parameters becomes necessary if the analysis at different lev-els is to be integrated.

● Exclusiveness of the levels of analysis. Unless the analysis at different levelsis done in a continuum, that is, within a given area, ecosystem, oragroecological zone, and by using related (common) parameters, its integra-tion and synthesis are extremely difficult.

● Unspecified definitions of rice environmental classes below the subecosystemlevel. There is no classification subdivision below the rice subecosystemlevel in any of the classification systems. Hence, different workers use dif-ferent terminologies, such as land management units, production units, riceenvironments, land-use units, etc. This is not a problem, however, if thereare equivalents and these can be specified. Also, from the review in thischapter, it has been noted that, at the higher levels of analysis, such as mega


and macro scales, rice environments are generally classified as ecosystemsand subecosystems, which are then characterized in terms of the prevailingbiophysical conditions using the key parameters and at the lower scales. Forexample, at the micro level, the analysis is simply a more detailed descrip-tion of the same, or similar, biophysical parameters. Such an analysis con-veys the same output (more detailed interpretation at the micro level) when,in fact, at different levels, it is expected to fulfill different objectives.

Some important parameters that are often overlookedThe following parameters are often overlooked in characterization.

● Groundwater table information. The groundwater information needed to de-termine the nature of water fluctuation, to interpret the duration and severityof drought, and to understand the groundwater contribution by capillary riseand its simulation is missing in most of the characterization studies reviewedin this chapter. This is probably because of the difficulty in gathering theinformation, which, by its own nature, is highly variable over time and het-erogeneous over space. However, rice yields are very sensitive to ground-water table depth, especially when it fluctuates within 1 m from the soilsurface (Wopereis 1993). Exclusion of this term may result in a 30% under-estimation of yields at 4 t ha–1 and a 90% underestimation at 1 t ha–1 (Boltonand Zandstra 1981).

● Surface hydrology. The information on sustained surface water depth, whichis highly dynamic, is also sparse in most of the studies reviewed in thischapter, probably for the same reason: the difficulty in gathering this infor-mation. Rice yields are also sensitive to surface flooding if it occurs at theactive tillering stage and the surface water depth patterns determine, to agreater extent, both the nature and severity of effects of hydrological stresson the crop and crop management practices.

● Seepage and percolation (S&P) rates. These are the main components ofwater balance and they strongly influence the presence or disappearance ofsurface water. In drought-prone areas, one of the reasons for frequent waterstress is high S&P rates, and this affects yields significantly. Using cropsimulation, Fukai et al (1995) estimated that a 2 mm d–1 reduction in S&P,from 6 to 4 mm d–1, would increase rice yields by more than 60% in Thai-land.

● Soil fertility mapping. Mapping units delineated in soil maps show polygonsof soil type/series and are based on soil genesis and classification. But exist-ing soil maps explain only 0–27% of the variances for the soil fertility pa-rameters that directly influence yield (Oberthur et al 1995). To be more rel-evant to increasing rice yields, technology development should focus onparameters that may limit yield (macro- and micronutrients, cation exchangecapacity, texture).

● Biotic stress profile. Biotic stresses (weeds, insect pests, diseases) are asimportant in rainfed systems as in the irrigated one. Savary et al (1997) showed

28 Singh et al

clearly that they are different in the two systems. Therefore, without under-standing biotic stresses, technology development may be addressing lessrelevant problems or may require modifications.

Conclusions

Characterization of rainfed rice environments for specific objectives is very usefulbecause it enhances research prioritization and technology development, delivery,and impact. Characterization for the sake of characterizing has limited value. Charac-terization without linkages to its higher or lower hierarchies also has limited value.Agronomic interpretations and developing relationships tremendously increase thevalue of characterization and mapping.

Characterization can be efficiently accomplished by first determining an objec-tive based on requirements, inventorying what is available and making use of it, andthen embarking on fresh data collection on new items when necessary. Adding infor-mation on the critical parameters that have been missing, such as hydrology, will addvalue to the already accomplished characterizations. Similarly, there is a need to es-tablish linkages over scales, disciplines, and institutions. Functional collaborationamong agencies seems to be an essential prerequisite. However, different institutionalinterests and sensitivities may be involved in data sharing, which one needs to beaware of in the collaborative efforts. A discussion on this among the partners from thebeginning may be quite useful and a clear outline of the responsibilities of and ben-efits to all involved is expected to enhance collaboration to serve the respective inter-ests. Several mechanisms exist, such as the Rainfed Lowland Rice Research Consor-tium and other consortia, and network projects, for exploring these opportunities tofurther enhance characterization and use it for productivity gains.

ReferencesAhmed NU, Magor NP, Islam S, Khan AH, Naseem SB. 1992. Extrapolation and testing of

technologies in Bangladesh. In: Proceedings of the 1990 Planning Workshop on Ecosys-tems Analysis for Extrapolation of Agricultural Technologies. Los Baños (Philippines):International Rice Research Institute. p 26-59.

Bolton F, Zandstra HG. 1981. A soil moisture based yield model of wetland rainfed rice. IRRIResearch Paper Series 62. Los Baños (Philippines): International Rice Research Insti-tute.

Buol SW, Cuoto W. 1981. Soil-fertility-capability assessment for use in the humid tropics. In:Greenland DJ, editor. Characterization of soils. Oxford (UK): Clarendon Press. p 254-261.

Conway G. 1986. Agroecosystem analysis for research development. Bangkok (Thailand):Winrock International.

FAO (Food and Agriculture Organization of the United Nations). 1977. FAO-UNESCO soilmap of the world. Vol. VII. South Asia. Paris (France): UNESCO.

FAO (Food and Agriculture Organization of the United Nations). 1979. FAO-UNESCO soilmap of the world. Vol IX. Southeast Asia. Paris (France): UNESCO.


FAO (Food and Agriculture Organization of the United Nations). 1988. Land resources ap-praisal of Bangladesh for agricultural development. BGD/81/035. Technical Reports1-7.

Fukai S, Rajatsasereekul S, Boonjung H, Skulkhu E. 1995. Simulation modeling to quantifythe effect of drought for rainfed lowland rice in Northeast Thailand. In: Fragile lives infragile ecosystems. Los Baños (Philippines): International Rice Research Institute.p 657-674.

Garrity DP. 1984. Asian upland rice environments. In: An overview of upland rice research.Proceedings of the Upland Rice Workshop, Bouaké, Côte d’Ivoire, 1982. Los Baños(Philippines): International Rice Research Institute. p 161-163.

Garrity DP, Agustin PC. 1984. A classification of Asian upland rice growing environments.Paper presented at the Workshop on Characterization and Classification of Upland RiceEnvironments, August 1984. Goiânia, Goiás (Brazil): CNPAF, EMBRAPA.

Garrity DP, Oldeman LR, Morris RA, Lenka D. 1986. Rainfed lowland rice ecosystems: char-acterization and distribution. In: Progress in rainfed lowland rice. Los Baños (Philip-pines): International Rice Research Institute. p 3-23.

Garrity DP, Agustin PC, Dacumos RQ, Pernito RN. 1992. A method for extrapolating rainfedcropping systems by land type. In: Proceedings of the 1990 Planning Workshop on Eco-system Analysis for Extrapolation of Agricultural Technologies, 22-25 May 1990,Tuguegarao, Cagayan, Philippines. Los Baños (Philippines): International Rice ResearchInstitute.

Garrity DP, Bruce RC. 1992. Rice ecosystems in Cambodia. Paper presented at the seminar onRemote Sensing and GIS in Agricultural Research, 20-22 July 1992. Los Baños (Philip-pines): International Rice Research Institute. 11 p. (In mimeo.)

Garrity DP, Singh VP, Hossain M. 1996. Rice ecosystems analysis for research prioritization.In: Evenson RE, Herdt RW, Hossain M, editors. Rice research in Asia: progress andpriorities. Los Baños (Philippines): CAB International and International Rice ResearchInstitute. p 35-58.

Higgins GM, Kassam AH, van Velthuizen HT, Prnell MF. 1987. Agricultural environments:characterization, classification and mapping. In: Bunting AH, editor. Proceedings of theRome workshop on agroecological characterization, classification, and mapping, 14-18April 1986. Wallingford (UK): CAB International. p 171-183.

Holdridge LR, Grenke WC, Hatheway WH, Liang T, Tosi JA. 1971. Forest environments intropical life zones: a pilot study. Oxford (UK): Pergamon.

Huke RE. 1982. Rice area by type of culture: South, Southeast, and East Asia. Los Baños(Philippines): International Rice Research Institute. 32 p.

Huke RE, Huke EU. 1997. Rice area by type of culture: South, Southeast and East Asia—arevised and updated database. Los Baños (Philippines): International Rice Research In-stitute. 59 p.

IRRI (International Rice Research Institute). 1984. Terminology for rice growing environments.Los Baños (Philippines): IRRI. 35 p.

IRRI (International Rice Research Institute). 1987. Annual report for 1986. Los Baños (Philip-pines): IRRI. p 432-436.

IRRI (International Rice Research Institute). 1989a. IRRI toward 2000 and beyond. Los Baños(Philippines): IRRI.

IRRI (International Rice Research Institute). 1989b. Program report for 1988. Los Baños (Phil-ippines): IRRI. p 430-436.

30 Singh et al

IRRI (International Rice Research Institute). 1990. Program report for 1989. Los Baños (Phil-ippines): IRRI. p 68-89.



IRRI (International Rice Research Institute). 1993. Program report for 1992. Los Baños (Phil-ippines): IRRI. p 62-66, 199-230.


Jones PG, Garrity DP. 1986. Toward a classification system for upland rice growing environ-ments. In: Progress in upland rice research. Proceedings of the 1985 Jakarta Conference.Los Baños (Philippines): International Rice Research Institute. p 107-116.

Kam SP, Tuong TP, Hoanh CT, Ngoc NV, Minh VQ. 2000. Integrated analysis of changes inrice cropping systems in the Mekong River Delta, Vietnam, by using remote sensing,GIS and hydraulic modeling. Poster presented at the XIXth ISPRS (International Soci-ety for Photometry and Remote Sensing) Congress, Amsterdam, The Netherlands, 16-23 July 2000.

Kassam AH, van Helthuizen HT, Higgins GM, Christoforides A, Voortman RL, Spiers B. 1982.Assessment of land resources for rainfed crop production in Mozambique.FAO:AGOA:MOZ/75/011, Field Documents 32-37. Rome (Italy): Food and Agricul-ture Organization of the United Nations.

KEPAS. 1985. The critical uplands of eastern Java: an agro-ecosystems analysis. KelompokPenelitian Agro-Ekosistem, Agency for Agricultural Research and Development, Re-public of Indonesia.

Koppen W. 1936. Gas geographische System der Klimate. In: Handbuch der Climatologie.Berlin (Germany): Borntrager.

Lightfoot C, Axinn N, Singh VP, Bottral A, Conway G. 1989. Training resource book foragroecosystems mapping. Los Baños (Philippines): International Rice Research Insti-tute and the Ford Foundation, India.

Lightfoot C, Singh VP, Paris T, Mishra P, Salman A. 1990. Training resource book for farmingsystems diagnosis. Manila (Philippines): International Rice Research Institute and In-ternational Center for Living Aquatic Resources Management.

Magbanua RD, Garrity DP. 1988. Acid upland agroecosystems: a micro-level analysis of theClaveria research site. Proceedings of the Acid Upland Research Design Workshop.International Rice Research Institute and Department of Agriculture. Region 10, Cagayande Oro City, Philippines. p 1-20.

Minh VQ. 1995. Use of soil and agro-hydrological characteristics in developing technologyextrapolation methdology: a case study of the Mekong Delta, Vietnam. M.S. thesis.University of the Philippines Los Baños, Los Baños, Philippines. 164 p.

Oberthur T, Doberman A, Neue HU. 1995. Spatial modeling of soil fertility: a case study inNueva Ecija Province, Philippines. In: Fragile lives in fragile ecosystems. Los Baños(Philippines): International Rice Research Institute. p 689-705.

Oldeman LR. 1980. The agroclimatic classification of rice-growing environments in Indone-sia. In: Cowell RL, editor. Proceedings of a symposium on the agrometeorology of therice crop. Los Baños (Philippines): International Rice Research Institute. p 47-66.

Oldeman LR, Frere M. 1982. A study of the agroclimatology of humid tropics of SoutheastAsia. Rome (Italy): FAO.


Papadakis J. 1975. Climates of the world and their agricultural potentialities. Buenos Aires(Argentina): Published by the author.

Savary S, Srivastava RK, Singh HM, Elazegui FA. 1997. A characterization of rice pests andquantification of yield losses in rice-wheat system in India. Crop Prot. 16(4):387-398.

Singh AN. 1987. Assessing extent of damage caused by flooding and drought in predominantlyrice cropland area using Landsat data. In: Proceedings of the Eighth Asian Conferenceon Remote Sensing, Jakarta, Indonesia. Tokyo (Japan): Asian Association of RemoteSensing. p 14:1-10.

Singh AN. 1988. Assessing impact of drought in a rainfed cropland area using Landsat data.Paper presented at the Ninth Asian Conference on Remote Sensing, 23-27 Nov. 1988,Bangkok, Thailand.

Singh RK, Singh VP, Variar M. 1994. Environmental characterization for on-farm research inrainfed uplands of Bihar, India: a case study. Rice Farm. Syst. Tech. Exch. 4:7-9.

Singh VP, Pathak MD. 1990. Rice growing environments in Bahraich district of Uttar Pradesh.Lucknow (India): Uttar Pradesh Council of Agricultural Research. 100 p.

Singh VP, Singh RK, Singh RK, Chauhan VS. 1993. Developing integrated crop-livestock-fishfarming system for rainfed uplands in Eastern India. J. Asian Farm. Syst. Assoc. 1(4):523-536.

Singh VP, Singh AN. 1995. Integration of different levels of rice ecosystems analysis. In: Frag-ile lives in fragile ecosystems. Los Baños (Philippines): International Rice ResearchInstitute. p 481-506.

Singh VP, Singh AN. 1996. A remote sensing and GIS-based methodology for the characteriza-tion and classification of rainfed environments. Int. J. Remote Sensing 17(7):1377-1390.

Singh VP, Sastri ASRAS. 1998. Sustaining rice farming in Eastern India: a conceptual andfactual analysis. Paper presented at the National Seminar on Farming Rice in EasternIndia, 19-22 March 1998. Lucknow (India): Uttar Pradesh Council of Agricultural Re-search.

Singh VP, Singh RK, Sastri ASRAS, Baghel SS, Chaudhary JL. 1999. Rice growing environ-ments of Eastern India: an agroclimatic analysis. Raipur (India): Indira Gandhi Agricul-tural University and Los Baños (Philippines): International Rice Research Institute.76 p.

Singh VP, Singh RK. 2000. Rainfed rice: best practices and strategies in eastern India. Manila(Philippines): International Rice Research Institute, Indian Council of Agricultural Re-search (India), International Fund for Agricultural Development (Italy), and Interna-tional Institute for Rural Reconstruction (Philippines). 292 p.

TAC (Technical Advisory Committee). 1990. Towards a review of CGIAR priorities and strat-egies. Progress report by the Technical Advisory Committee (TAC/CGIAR). Washing-ton, D.C. (USA): Consultative Group on International Agricultural Research.

Thornthwaite CW. 1948. An approach towards a rational classification of climate. Geog. Rev.38:55-94.

Tuong TP, Hoanh CT, Khiem NT. 1991. Agro-hydrological factors as land qualities in landevaluation for rice cropping patterns in the Mekong Delta of Vietnam. In: Deturck P,Ponnamperuma FN, editors. Rice production on acid soils of the tropics. Kandy (SriLanka): Institute of Fundamental Studies. p 23-30.

WARDA (West Africa Rice Development Association). 1992. Annual report for 1991. Bouaké(Côte d’Ivoire): WARDA.

WARDA (West Africa Rice Development Association). 1993. Annual Report for 1992. Bouaké(Côte d’Ivoire): WARDA.

32 Singh et al

Widawsky DA, O’Toole JC. 1990. Prioritizing the rice technology research agenda for EasternIndia. New York, N.Y. (USA): Rockefeller Foundation.

Wopereis MCS. 1993. Quantifying the impact of soil and climatic variability on rainfed riceproduction. Ph.D. thesis. Wageningen Agricultural University, Wageningen, The Neth-erlands.

NotesAuthors’ address: International Rice Research Institute, DAPO Box 7777, Metro Manila,

Philippines.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-

acterizing and understanding rainfed environments. Proceedings of the InternationalWorkshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999,Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute.488 p.

Characterizing environments for sustainable rice production 33

Characterizing environmentsfor sustainable rice productionVan Nguu Nguyen

Growth in the world’s rice production has slowed down. Since 1990, thegrowth rate of rice production has been lower than that of the population.This indicates the need to increase efforts to improve rice productivity andbring more land area under rice cultivation. Substantial efforts have beenmade to characterize and classify the environments for agricultural produc-tion in general and for rice production in particular during the past threedecades. Major efforts in the characterization and classification of theseenvironments are reviewed and selected examples of the contribution of theseexercises to rice production via the development of rice technologies, expan-sion of rice area, transfer of rice technologies, and others, such as the as-sessment of investments in agricultural research, are provided and discussed.

Rice production factors such as varieties, water and land/soil resources,insects and diseases, socioeconomic issues, and global climate have alsoevolved substantially during the past 30 years. A new generation of rice vari-eties with higher yield potentials and better resistance to abiotic and bioticstresses has been developed using hybrid rice and biotechnology. The in-creasing deficiency of nutrient elements and pressure from insects and dis-eases have been observed in intensive rice production systems. Socioeco-nomic factors such as labor availability and wages and availability of inputsand credits have changed with improvements in the economies of rice-pro-ducing countries. The global climate has been warming and discussions coverthe implications of these changes for the suitability of rice production underdifferent environments.

The globalization of the world economy and the decline in public invest-ments for agricultural research activities require that future efforts in charac-terization and classification be evaluated in terms of returns to agriculturalproduction. Discussions refer to the issues affecting the efficiency and appli-cability of characterizing and classifying the environments of rice productionsuch as universality, completeness, objectivity, and scales and costs of theexercises.

34 Van Nguu Nguyen

Rice is the staple food crop for more than half the world’s population. Its popularityhas increased steadily, not only in rice-eating countries but also in countries wheretraditionally it is not an important food crop. The slowdown in the growth of riceproduction is serious because of the continuing growth of population. Sustainablerice production in the near future therefore requires more efforts to improve rice pro-ductivity and to bring more land area into rice cultivation. During the past three de-cades, substantial efforts have been made to characterize and classify the environ-ments of agricultural production, especially the production of important food cropssuch as rice. The results of these exercises have been applied in developing cropproduction technologies, expanding new crop areas, transferring technologies, andothers. Recently, results of characterization and classification of environments havealso been applied in assessing investments in agricultural research.

Rice yields are affected by variety, ecological conditions during the growingseason, and socioeconomic factors that affect farmers’ crop management. Rice cropsyield highest when planted in the best-suited environments. Factors affecting riceproduction, however, have evolved much during the past three decades and need tobe included in future efforts to characterize and classify the environments of riceproduction. In the competitive markets created by the globalization of the worldeconomy, improving rice productivity and expanding rice areas must be done in themost efficient manner. Public investments in rice research have also been declining.In the future, therefore, characterizing and classifying the environments of rice pro-duction must take into consideration the following areas: universality, completeness,objectiveness, and scales and costs.

The challenge to the world’s rice production

Rice is the world’s most important food crop. In 1997, about 2.9 billion people de-pended mainly on rice for food calories and protein. The popularity of rice has alsoincreased in Africa and Latin America, where traditionally rice has not been an im-portant food crop. The worldwide annual growth rates of population and rice produc-tion, harvested area, and yield since 1970 (Table 1) show that world rice productionhas increased continuously, but at varying growth rates. The annual growth rate was2.7% in the 1970s, 3.1% in the 1980s, and 1.3% in the first half of the 1990s. Acomparison between the growth rates of rice production and population since 1970shows that, for the first time since 1990, rice production has grown more slowly thanpopulation.

During the 1970s, the high annual growth rate of rice production was caused byboth a high increase in yield and a moderate increase in rice area, whereas the rapidgrowth in rice production during the 1980s came principally from improvements inrice productivity. Rice yield grew 1.8% annually in the 1970s, 2.8% in the 1980s, andonly 1.1% in the first half of the 1990s, whereas the annual growth rate of harvestedrice area decreased from 0.8% during the 1970s to 0.2% in the 1980s (Table 1).

The trend of reduced growth of rice harvested area indicates that future riceproduction increases will come mainly from improvements in productivity unless


major development activities are undertaken to bring more land under rice cultiva-tion. The very low annual growth rate of rice yield observed since 1990 is therefore acause for concern and it has been the topic of numerous reviews (Pingali and Rosegrant1994, Cassman and Pingali 1995, Pingali et al 1997). Regardless of food consump-tion trends, the slowdown in growth of rice production is particularly serious consid-ering the continuing growth of population. This declining growth trend in rice pro-duction needs to be reversed if the world’s rice production is to meet popular demand.This requires efforts to increase rice yield, expand rice area, or a combination of both.Historical evidence indicates that the world’s efforts in these directions could be ef-fectively assisted through vigorous characterization of the environments of rice pro-duction.

Characterizing environments and their impacts on rice production

Chandler (1979) considered the inventory/analysis of natural resources as an essen-tial element for a successful national rice program. Rice is grown from about 50°N toabout 35°S, from below sea level to about 2,700 m above, and in a wide range ofecological conditions, from dry land where soils are freely drained to flooded landwith the depth of flooded water reaching several meters. The development of irriga-tion further modified the ecological environments of rice production. Human inter-vention in rice production increases the complexity of the already diverse rice pro-duction environments.

Several systems have characterized and classified the environments of rice pro-duction. These activities in characterization and classification have provided the ba-sis for increasing rice production by developing improved technologies, developingnew rice areas, transferring improved technologies, and making other investmentsaimed at supporting rice production.

Major activities in characterizing and classifying environmentsEarly attempts to characterize and classify the environments of rice production. Ricescientists and researchers from different national and international institutions havemade considerable efforts to classify and characterize the environments of rice pro-duction. In late 1973, the International Rice Research Institute (IRRI) organized its

Table 1. Annual growth rates (%) of the world’s population and riceproduction, harvested area, and yield.

Rice

Period Population Production Harvested Rice yieldarea

1970-79 2.03 2.71 0.80 1.761980-89 1.86 3.14 0.23 2.801990-96 1.55 1.31 0.23 1.10

36 Van Nguu Nguyen

first working group to identify the macro soil/climate zones of rice production inSoutheast Asia (IRRI 1974). A comprehensive agroclimatic classification for evaluat-ing cropping systems in Southeast Asia based on the amount of monthly rainfall andlength of consecutive wet months was reported by Oldeman in 1974. A wet month isdefined as a month with at least 200 mm rainfall. The agroclimatic environments ofSoutheast Asia were classified into four zones (Oldeman 1974):

● Zone I: more than 9 consecutive wet months● Zone II: 5–9 consecutive wet months; zone II has four subzones● Zone III: 2–5 consecutive wet months; zone III has three subzones● Zone IV: less than 2 consecutive wet monthsSince then, several attempts to characterize and classify rice production envi-

ronments have been carried out by different scientists working in different locationsaround the world. The bases for characterizing rice production environments used upto the mid-1980s, however, are different (Table 2). Recognizing the importance ofhaving a common agreement on environmental terminology so that plant type can bebetter related to environments, IRRI established in 1982 an international committeeto develop an agreed-upon terminology and classification for rice (IRRI 1984).

The recent characterization and classification of the environments of rice pro-duction in inland valleys in West Africa. Rice is an important food crop in West Af-rica, but local production has not been able to meet popular demand, resulting in thespending of large amounts of foreign exchange by governments in the region to im-port rice. The region has large wetland areas in inland valleys, however, that are stillnot fully exploited. Considerable efforts have therefore been made recently to charac-terize and classify the environments in these inland valleys, where rice has been atraditional crop. The efforts have been carried out since 1982 by various national andinternational institutions, such as the International Institute of Tropical Agriculture(IITA) (Windmeijer and Andriesse 1993), Conférence des Responsables de Recher-che Agricole Africains (CORAF) (Albergel et al 1993), the Centre de CoopérationInternationale en Recherche Agronomique pour le Développement (CIRAD) (Legoupiland Bidon 1995), and the West Africa Rice Development Association (Becker and

Table 2. The classification of rice production environmentsduring the 1970s to early 1980s.

Basis of classification/characterization Exercises (no.)

General surface hydrology 15Physiographic source of water 3Landform and soil units 3Matrix of ecological factors 9Soil suitability 3Crop season, intensity, and management 3Comprehensive (combination of above 6 items) 2Total 38

Source: Bowles and Garrity (1988).


Diallo 1992). In 1990, the FAO Regional Office in Accra, Ghana, established a Tech-nical Co-operation Network in Wetland Development and Management (WEDEM)aimed at promoting the environmentally sound development and management ofwetland resources for sustainable food production and exchange of information. Thecharacterization and classification of the agrosystems in inland valleys in West Africaby the earlier-mentioned institutions were based on physical (e.g., climate, lithology,landform, soil, and hydrology), biotic (e.g., vegetation, insects, diseases, and weeds),and management (land use, distance from farm to market) factors. Land use wasdescribed by socioeconomic parameters such as labor, capital input, and manage-ment.

The FAO’s agroecological zoning for agricultural production. Recognizing theimportance of classifying and characterizing natural resource bases, the Food andAgriculture Organization of the United Nations (FAO) began in 1976 to develop theagroecological zones (AEZ) methodology (FAO 1976) and supporting databases andsoftware packages to provide solutions for land resources analysis in member coun-tries, linking land-use outputs with other developmental goals in areas such as foodproduction, food self-sufficiency, cash crop requirements, and population-supportingcapacity. The key elements of AEZ are based on the FAO’s Framework for LandEvaluation (FAO 1976), which emphasizes the need to characterize land-use types(LUTs) as a necessary precursor to land evaluation and land-use planning. Climatic,soil, and plant parameters were used to calculate the length of the growing period(LPG) for various crops and to determine crop suitability. Land productivity is esti-mated at two levels of input application: the minimum and the optimum. The first useof the AEZ methodology was to assess the production potential of land resources inthe developing world based on climatic data and the 1:5,000,000 scale FAO/UNESCOSoil Map of the World.

The AEZ methodology was applied entensively to evaluate the suitability andland productivity under rainfed conditions of five major food crops—wheat, rice,maize, barley, pearl millet, and sorghum; three root crops—white potato, sweet po-tato, and cassava; two leguminous crops—soybean and Phaseolus bean; and othercash crops in 117 countries in Africa, Central and South America, and Asia. One ofthe major products of this exercise was the mapping of zones suitable for producingvarious food and cash crops under rainfed conditions in the developing countries ofAfrica, Central and South America, and Asia, an example of which appears in Figure1. Although maps were produced for other crops, however, none was made for rice(FAO 1978, 1980, 1981), perhaps because of the uniqueness of the rice-growing eco-logical environments. Outside of upland ecologies, flooding and its patterns mostlydominate rice environments.

The methodology and findings of the AEZ methodoloy were presented at aFAO conference in 1983, which, recognizing the importance of such work for devel-opment, recommended that similar work be undertaken at the national level (Antoine1994). Since then, FAO has assisted several countries in learning the methodologyand applying and adapting it to tackle issues of land, food, and people at the nationaland subnational level. FAO’s agroecological zone models were successfully linked

38 Van Nguu Nguyen

Fig. 1. Agroclimatic suitability assessment for rainfed pearl millet production in South America(after FAO 1981).

with geographic information systems (GIS) to appraise natural resources to supportthe population in Kenya, Nigeria, and the Sahel countries in West Africa (Antoine1994).

Selected examples of the contribution of the characterizationand classification of environments to sustainable rice productionExample 1. Developing cropping systems to intensify rice production in rainfed low-lands. Rice-cropping intensification—or the growing of two or more rice crops on the

TropicsSubtropics

Temperate

Summer/winter rainfall

Normal isoline

Intermediate isoline

High altitudes/cold temperatures

High altitudes/cool temperatures

Very suitable

Marginally suitable

Not suitable

Suitable


same piece of land in a year—increases not only rice production but also farmers’incomes and employment opportunities. After the development of short-growth-du-ration and photoperiod-insensitive rice varieties, cropping intensification has beenwidely practiced in irrigated areas under tropical climate. On the other hand, the tra-ditional method of growing rice in rainfed lowlands allows farmers to grow only onerice crop per year during the wet season.

The results of the agroclimatic classification mentioned earlier (IRRI 1974,Oldeman 1974) led to a study at IRRI aimed at developing new rice-based croppingsystems for the rainfed lowland rice areas in the early 1970s. With the encouragingresults of IRRI’s study, the Philippine Council for Agriculture and Resources Re-search and Development agreed in 1975 to join with IRRI to implement a cooperativeapplied research project on rainfed cropping systems. The project aimed at develop-ing, evaluating, and disseminating improved rice-based cropping systems for therainfed lowland rice areas in the Philippines. The most significant result of this coop-erative project was the development and transfer of a new cropping system that al-lowed the growing of two rice crops per year in rainfed lowland areas where the wetseason is at least 6 mo with rainfall of 200 mm or more (Fig. 2).

The Philippines’ national average rice yield before implementation of the projectwas about 1.2 t ha–1. Adopting this new cropping system enabled farmers to grow tworice crops and an upland crop per year in many rainfed lowland areas, with a possibleannual yield of 9–10 t ha–1 for rice (Cardenas et al 1980). The higher rice productionmultiplied the incomes of farmers several times.

With the impressive results obtained in the Philippines, many rice productionprograms in Asia have adopted and then modified this new cropping system and trans-

Fig. 2. Cropping calendar of introduced and traditional cropping sys-tems in some rainfed lowland rice areas in the Philippines (afterCardenas et al 1980).

700

600

500

400

300

200

100

0J F M A M J J A S O N D

Month

Rainfall (mm)

Introducedsystem

Traditionalsystem

Firstcrop

Secondcrop

Normaltransplanting

Direct seedingHarvestingTransplantingSeedbed preparation

40 Van Nguu Nguyen

ferred it for wide adoption by farmers for increasing both rice production and theproduction of other food crops and farmers’ income. The environments of rice pro-duction in many Asian countries have zones with agroecological conditions similar tothose in the Philippines where this new cropping system was tested (Oldeman andFrere 1982, Huke 1982a,b).

Example 2. Expanding rainfed lowland and irrigated rice areas in West Africa.West Africa has abundant wetland areas, but traditional rice production was carriedout mainly in upland ecologies. As mentioned earlier, activities to characterize andclassify the environments in inland valleys in West Africa were strengthened startingin 1982. The rapid increase in harvested rainfed lowland and irrigated rice area inWest Africa after 1985 (Table 3) could be partially attributable to using the results ofthe work on characterization and classification of the environments in inland valleysin the region. In 1975, the harvested area from rainfed lowland rice was only 441,000ha. It increased to 516,000 in 1985, an increase of about 17% after 10 years. In 1995,it was 979,000 ha, an increase of about 90% after 15 years. A similar trend in areaexpansion was also observed in irrigated rice (Table 3). The expansion in rice areahas greatly contributed to the rapid growth of rice production in the region.

Example 3. Guidelines for fertilizer recommendations in rice production inBangladesh. Rice is the dominant food crop in Bangladesh’s agriculture and is thestaple food for more than 120 million Bangladeshis. Intensification of rice productionhas resulted in higher demand for fertilizer because of higher crop removal. This,coupled with the imbalance in fertilizer application, mainly nitrogen, has led to in-creasing deficiencies of nutrient elements such as phosphorus, potassium, zinc, andsulfur in many rice soils, thus endangering the sustainability of national rice produc-tion. The Bangladesh Agricultural Research Council began and implemented a projectaimed at applying agroecological zoning to support rice production during the early1980s, with financial and technical support from the United Nations DevelopmentProgramme (UNDP) and FAO. The project’s activities resulted in “a fertilizer recom-mendation guide” for extension workers to use in planning their activities and assist-ing rice farmers. This has greatly contributed to the sustained growth of rice produc-tion in the country (Karim 1994). Table 4 shows part of the fertilizer recommendationguide for rainfed rice-cropping systems.

Other examples of impact of the classification and characterization of environ-ments. Advances in geographic information systems (GIS), crop simulation model-

Table 3. West Africa’s harvested rice area, 1975 to 1995 (000 ha).

Year Total Irrigated Rainfed Upland Otherslowland

1975 2,292 82 441 1,379 4011980 2,561 171 507 1,574 3101985 2,706 231 516 1,599 3701990 3,380 337 753 1,916 3761995 3,886 492 979 1,983 432


Table 4. Selected fertilizer recommendation guide for rainfed rice-based cropping systems inagroecological region 11 in Bangladesh: High Ganges River floodplain.

Fertilizer recommendationCropping pattern (kg ha–1)

Land and soil characteristicsa

Season Crop N P2O5 K2O S Zn

Land type: highland; organic Rabi Wheat 60 30 30 10 –matter: low; pH: 6.1–7.9; Kharif 1 B. aus (L)b 30 – – – –texture: silt-loam, K-bearing Kharif 2 Fallow – – – – –minerals: medium

Rabi Mustard 80 60 40 30 3Kharif 1 Groundnut/jute 40 – 10 – –Kharif 2 B. aus (L) 30 – – – –

Rabi Chickpea/lentil 20 40 30 10 –Kharif 1 B. aus 30 – – – –Kharif 2 Fallow – – – – –

Land type: medium highland; Rabi Khesari 10 20 – – –OM: low; pH: 6.1–7.9; Kharif 1 Aus + aman (L) 30 20 20 – –texture: loamy; K-bearing Kharif 2 – – – – –minerals: medium

Rabi Chickpea 20 50 30 10 5Kharif 1 B. aus (L) 30 20 20 – –Kharif 2 T. aman (L) or 50 20 20 – –

T. aman (HYV) 70 20 20 10 –

Rabi Wheat 60 30 30 10 3Kharif 1 Jute 45 – 10 – –Kharif 2 T. aman (L) or 50 20 20 – –

T. aman (HYV) 70 20 20 10 –

Land type: medium lowland; Rabi Khesari 10 30 – – –OM: medium; pH: 6.1–7.9; Kharif 1 Aus + aman (L) 30 – – – –texture: clayey; K-bearing Kharif 2 – – – – – –minerals: high

Rabi Khesari 10 30 – – –Kharif 1 B. aman (L) 30 – – – –Kharif 2 – – – – – –

Rabi Boro (L) 60 40 20 10 3Kharif 1Kharif 2

aOM = organic matter. bB = broadcast, T = transplanted, L = local variety, HYV = high-yielding variety.Source: Karim (1994).

42 Van Nguu Nguyen

ing, remote sensing, and computing technologies have further strengthened the appli-cability of AEZ in agricultural planning and development. Many countries have usedthe AEZ methodology (FAO 1994) in assessing land productivity and population-supporting capacity; in land evaluation and land-use planning; in assessing environ-mental degradation due to agricultural production; and in research planning, technol-ogy transfer, farming systems analysis, and recommendations for input applicationand supply. Detailed information on these applications was published in the FAO’sWorld Soil Resources Report Number 75 (FAO 1994).

Application of the characterization and classification of the environments ofagricultural production has recently been extended to quite a new area: the evaluationof investments in agricultural research. The methodology for evaluating alternativeresearch investments calls for assessing the impacts of past research on production(ex post analysis) and the possible or potential impact of the currently proposed re-search alternatives (ex ante analysis). Characterizing and classifying the environments,especially those that employ the AEZ methodology, can provide quantified inputs tosuch assessment (Pardey and Wood 1991, Wood and Pardey 1993). Dividing geo-graphic space into AEZs provides an estimate of the area that could benefit from theresults of the proposed research investments. The homogeneous conditions undereach AEZ, on the other hand, facilitate quantifying the response to (or outputs of) theapplication of new technologies resulting from the proposed research investments.Research investments could then be disaggregated to commodities, subtypes, envi-ronments, and problem and discipline domains.

For example, investments in rice research can be grouped into irrigated, rainfedlowland, upland, and deepwater and tidal wetland, and then into genetic improve-ment, crop management, and crop protection. The products of the research invest-ments could be further classified into non-site-specific (applied equally to all AEZs),site-specific (applied to only one AEZ), or multivariable site-specific (variable AEZs).

Rice production factors as guidelines for environmental characterization

Variety, ecological conditions during the growing season, and the socioeconomic fac-tors that affect farmers’ crop management determine the yield of a rice crop. Thefactors affecting rice production have undergone substantial evolution during the past30 years. This requires more attention to the classification and characterization of theenvironments of rice production in the future.

Rice varietal developmentThe rapid growth in world rice production during the 1980s (Table 1) came mainlyfrom the gain in productivity. The results of the maximum yield studies carried out byIRRI during the 1970s showed that, in tropical climate areas, the potential yield ofrice varieties was about 3.7–6.8 t ha–1 in irrigated ecologies, 2.5–4 t ha–1 in rainfedlowlands, and 2 t ha–1 in upland and other ecologies (Table 5). Cassman et al (1997)reported that yields of 4 to 5 t ha–1 are normally obtained in irrigated areas in severaltropical countries. The average yields in 1997 in major rice-producing countries such


as Bangladesh, Brazil, India, Myanmar, Nigeria, the Philippines, Thailand, and Viet-nam were still below 4 t ha–1, indicating the limited improvement in rice productivityin other ecologies.

In 1997, about 54% of the world’s rice harvested area came from irrigated ecolo-gies, 30% from rainfed lowland ecologies, 11% from upland ecologies, and 5% fromother ecologies such as deepwater and tidal wetlands or mangroves. Irrigated rice wasresponsible for about three-quarters of the world’s total rice production, indicatingthe limited contribution of rice production in other ecologies. This further confirmsthe observation on the limited success of activities aimed at improving rice produc-tivity in rainfed environments during the past 30 years.

The yield potentials of high-yielding rice varieties (HYVs) in tropical areas,however, have not improved further after the development of IR8 in the late 1960s,although yield efficiency of rice varieties has been improved with the development ofearly maturing HYVs. Increasing efforts have therefore been made to develop newrice varieties with higher yield potentials. Since 1982, the Japanese government hasbeen promoting a project to develop super-high-yielding varieties with the target ofincreasing rice yield by 50% in 15 years based on wide crosses between indica andjaponica varieties. Breeding for the new plant type that could increase the yield po-tential of tropical rice by 25% to 50% began at IRRI in 1989. Tropical japonica vari-eties have been used as sources for desirable traits in this project. After learning of thesuccessful application of this technology for increasing rice production in China, FAO,IRRI, and several other national institutions have been promoting the developmentand use of hybrid rice (Tran and Nguyen 1998). The West Africa Rice DevelopmentAssociation has made considerable efforts to develop rice varieties for low-inputmanagement areas in West Africa from crosses between Oryza sativa and O.glaberrima. Therefore, new and more vigorous characterization and classification of

Table 5. Estimated maximum farm yields (t ha–1) for different types of riceland in 11 Asian countries.

Country Irrigated Supplemental Rainfed Upland anddry-season irrigated wet-season lowland deepwater

Philippines 5.9 4.6 3.5 2.0India 6.8 5.4 4.0 2.0Indonesia 5.9 4.8 3.6 2.0Thailand 4.4 3.7 2.5 2.0Bangladesh 6.6 4.9 3.7 2.0Vietnam 5.8 4.1 3.1 2.0Sri Lanka 5.7 5.3 4.0 2.0Myanmar 6.0 4.8 3.6 2.0Pakistan 6.0 – – –Nepal – 4.8 3.6 2.0Malaysia 6.0 4.8 3.6 2.0

Source: Chandler (1979).

44 Van Nguu Nguyen

rice environments will be needed to develop management techniques to allow newrice varieties to fully express their potential.

Biotechnological tools have been increasingly used to develop rice varietieswith better resistance to insects, diseases, and abiotic stresses. Transgenic rice plantswith Bt genes have been created for stem borer resistance. Herbicide-resistanttransgenic rice varieties have also been developed. Substantial progress has been madein the use of biotechnological tools for creating drought- and salinity-tolerant ricevarieties (Khush et al 1999). These developments may influence the classification ofthe suitability of rice varieties for production under different environments.

Water suppliesRice depends on water for normal growth and development. An insufficient watersupply leads to drought stress, whereas an oversupply of water results in completesubmergence. Rice yields under both situations are usually low. The water supply inrice production is controlled best under irrigation systems. Irrigation water is increas-ingly becoming less and less available for rice production in many countries becauseof the depletion of aquifers, salinity, and competition for water from urbanization andindustrialization. In several other countries, the high costs of irrigation infrastructurehave limited the building of new irrigated rice schemes. In the near future, therefore,large rice areas will remain under rainfed conditions. Considerable deepwater riceareas in Bangladesh, Cambodia, and Vietnam have been converted for growing boroand/or dry-season irrigated rice. Sizeable rice areas, however, still suffer from fre-quent deep floods.

Drought. Rice plants generally have shallow root systems; therefore, water de-ficiency or drought stress has been considered to be the most important yield-limitingfactor in rainfed lowland and upland rice. It is also a major constraint to rice produc-tion in most deepwater rice areas and in some irrigated and tidal wetland areas. Underrainfed conditions, water deficiency can occur at any time during the cropping sea-son, but especially during the early and late stages of the crop. The degree of damageto rice crops depends on the time in relation to the development stage of the riceplants when water deficiency occurs and its intensity and duration. The damage isusually heavy and irreparable when intensive water deficiency occurs during the re-productive and flowering stages. Yield losses of 1 t ha–1 or more may occur after 10 dof continuous water deficiency during these stages. Some rice varieties exhibit thecapacity to fully recover and resume normal growth after being exposed to droughtstress during the vegetative stages, but their growth duration is generally prolonged.This could make rice varieties unsuitable in areas or zones that were originally classi-fied as suitable based on plant growth duration.

The main sources of the water supply to rainfed lowland rice in undulatingterrain are rainfall, interflows, and streams activated by local rain. In basins, deltas,estuaries, and lake fringe, the groundwater table may rise during the rainy season andcome within easy reach of the rice roots. Under rainfed conditions, therefore, thewater supply to wetland rice fields depends on both rainfall and the hydrology of therice field. Farmers often carry out bunding and leveling of rice fields to conserve


water for rice production. The feasibility of field bunding and leveling depends greatlyon the topography of the area. Furthermore, the soil’s capacity to retain water has animportant influence on the water supply for rainfed rice during its growing season.Therefore, parameters such as rainfall and its distribution, hydrology of rice fields,topography, and soil water retention capacity should be used when estimating theperiod of water deficiency and its duration.

Complete submergence. Although rice plants are well known for their ability totransport oxygen from the air into their root systems, flooding, with consequent cropsubmergence, may severely damage the rice crop. Most wetland rice varieties, in-cluding deepwater ones, can withstand complete submergence for at least 6 d before50% of the crop dies, whereas 100% mortality occurs in all varieties within 14 d ofcomplete submergence (Setter et al 1995). The growth and development of rice plantsunder complete submergence are also affected by the quality of floodwater. In addi-tion to direct damage to rice crops from submergence, stagnant water causes exces-sive soil reduction, which alters the chemistry of wetland soils and usually causesnutrient deficiency or toxicity or both to the rice crop. Rice yields from fields whereprolonged stagnant flood occurs are usually low. Susceptibility to complete submer-gence therefore needs to be considered when we characterize rice production envi-ronments.

Adverse soil conditionsA decline in rice yield in long-term intensive rice production systems has been ob-served, at least at the experimental level (Cassman et al 1997). At the farm level, adecline in total factor productivity has been observed. Farmers have to apply moreand more production inputs to obtain the same rice yield (FAO 1997). Undoubtedly,soil physicochemical and biological environments change under continuous submer-gence for a long period. Deficiencies in phosphorus, potassium, zinc, and sulfur haveexpanded in many lowland rice areas (Cassman et al 1997). Increased salinity in ricesoils under intensive and continuous irrigated rice cultivation has also been increas-ingly observed (Pingali and Rosegrant 1994). These parameters need to be well quan-tified and characterized if solutions to the reversing of declining trends are to befound.

In Asia, land areas that are suitable to rice production have been used. Substan-tial wetland areas in coastal plains in some countries, however, are still available andcould be developed for rice production. Many of these land areas, however, are influ-enced by tidal water and have unfavorable soil conditions such as salinity, acidity,and peat. A high level of iron concentration is a major constraint to yield in manylowland rice areas, especially the inland valleys in sub-Saharan Africa, where about140 million ha of this type of land are still available (Ton That 1982). Iron toxicity hasbeen cited as a yield constraint to wetland rice in Brazil (Pulver, personal communi-cation). The characterization and classification of rice production environments thattake into account these parameters would be useful, especially for selecting areas forexpanding rice production.

46 Van Nguu Nguyen

Insects and diseasesInsects and diseases reduce rice yield. Intensification in rice production has changedthe types and pressures of many insects and diseases. Although insects and diseasescan be controlled with appropriate management techniques, their pressures need to becharacterized and classified, at least for varietal improvement activities. The use ofresistant varieties is still the cheapest and most effective measure for controlling in-sects and diseases in rice.

Socioeconomic issuesAlthough an area may be suitable for rice production based on ecological environ-ments, rice production may not necessarily be the best suited based on socioeconomics.Many areas suitable to rice production in southern Brazil, Argentina, and Venezuelaare either planted to other crops or under fallow due to the high costs of rice produc-tion and irrigation infrastructure. With the globalization of the world economy, riceproduction needs to be evaluated for its competitiveness not only with rice produc-tion in other areas but also with the production of other crops in the same area.

The costs of providing favorable conditions for rice production such as irriga-tion infrastructure, an adequate supply of inputs and credit, and better marketing ofrice need to be included in characterization. It is well known that improved rice tech-nologies, such as high-yielding varieties, could not make a significant contributionwithout the adequate availability of production inputs. The yield of IR8 does notdiffer significantly from that of its parent, Peta, unless fertilizers are applied.

The labor force for rice production in many countries has been shrinking rap-idly due to a combination of population control, improvement in income, and em-ployment opportunities created to improve national economies. Also, in many coun-tries, the population of rice farmers has become older and older because of the migra-tion of young people to urban centers and job preferences. The scarcity of labor andhigh rural wages have led to a shift in the application of rice technologies in thesecountries. Land preparation, harvesting, and threshing operations have become moreand more mechanized, direct seeding has increasingly replaced transplanting as themain crop establishment method, and chemical weeding has become more popularthan hand weeding. In countries such as India, the Philippines, and Thailand, womenhave been participating more and more actively in rice production operations.

Socioeconomic factors vary with changes in the national economy. Understand-ing socioeconomic factors could have an important bearing on activities aimed atdeveloping new rice land and at creating the appropriate conditions for adopting im-proved technologies. Socioeconomic factors were considered as major constraints tothe agricultural development of inland valleys in West Africa (Table 6).

Global climate changeGlobal climate change has been increasingly observed. The global climate has warmedbecause of the emission of CO2 and other gases such as methane and nitrite oxide andtheir accumulation in the atmosphere. Temperature and CO2 concentration are impor-tant parameters of the photosynthetic pathway, whereas temperature influences respi-


ration. A change in temperature regime may also lead to changes in rainfall and itsdistribution and cloudiness and its distribution, and thus solar radiation, wind speed,and biological activities in the soil-air-plant system. These factors have large effectson rice growth, development, and yield.

Major issues in characterizing rice environments

Experiences gained in characterizing and classifying rice production environments inthe past could provide insights for future exercises in this domain. The following aresome of the issues that future activities in characterizing rice production environ-ments may have to consider.

The universality and compatibility of the systems usedTable 2 shows that the bases used in many systems for classifying and characterizingrice production environments before the mid-1980s were not similar. This lack ofuniversality has limited the usefulness of environmental classification and character-ization, at least in terms of collaboration among rice research institutions and scien-tists to develop and transfer rice technologies. This was perhaps the main reason forIRRI to establish the International Committee on Terminology of Rice Growing En-vironments in 1982.

Table 6. Major constraints to agricultural devel-opment of inland valleys in West Africa.

Constraint Rankinga

Weeds 4.7Lack of water control 4.7Lack of inputs 4.6Labor shortage 4.4Crop diseases 4.3Lack of credit 4.3Nutrient status 4.2Land tenure 4.1Poorly adapted varieties 3.9Soil erosion 3.7Soil structure 3.6Insects/pests 3.6Marketing 3.6Iron toxicity 3.4Acidification 3.2Land clearing 3.2Human health 3.1

aOn a scale of 1–5, where 1 = not important and 5 =very important.Source: Jasmin and Andriesse (1993).

48 Van Nguu Nguyen

The completeness of characterization and classificationIncompleteness in characterization was another weakness of many early attempts tocharacterize and classify rice-growing environments. Several environmental factorsmay be static, but several other parameters are very dynamic in nature. For example,although environmental parameters such as altitude, longitude, latitude, soil texture,landform and slope, and even soil pH are more or less static, parameters such asgroundwater table, depth of flooded water layer, temperature, solar radiation, andothers are very dynamic in nature. An incomplete characterization may lead to imper-fect classification. For example, theoretically, rainfed rice production is not feasiblein areas where rainfall is less than 800 mm. Rainfed rice production has been success-ful, however, in many inland valleys in the Sudan Savanna Zone in West Africa,where annual precipitation ranges from 550 to 900 mm (Windmeijer and Andriesse1993). The water supply in rice production in this area comes not only from rainfallbut also from the groundwater table, runoff, and seepage from surrounding areas.

The concept of minimum data set (MDS) has been proposed for characterizingthe ecological conditions of crop production. This concept has been used in manyprojects, such as the project on the International Benchmark Sites Network forAgrotechnology Transfer (Uehara and Tsuji 1991). The MDS is supposed to providebasic environmental factors that have potential effects on plant growth, health, anddevelopment. The MDS, however, may be too simple for some studies. For example,results of experiments to evaluate rice germplasm for salinity tolerance may be mis-leading if soil alkalinity was not included in the environmental characterization of theexperimental sites. Soil salinity is usually associated with soil alkalinity and bothhave negative effects on the growth and development of rice plants. Similarly, Pen-ning de Vries et al (1989) considered that the minimum data sets used in many experi-ments to evaluate drought tolerance of rice varieties are not adequate for simulationmodeling. They opined that, to evaluate the level of drought stress in a given environ-ment, parameters such as rainfall, humidity, light, temperature, and soil characteris-tics in the root zone are needed. Andriesse and Fresco (1991) noted the weakness ofthe broad agroecological classification of rice environments when they reviewed theresults characterizing rice-growing environments in West Africa.

The objectiveness of characterization and classificationThe sources of information used to characterize and classify rice production environ-ments are another factor. Most activities to characterize and classify rice productionenvironments in the past were carried out mainly by researchers and scientists orpeople responsible for rice development. The characterization and classification maytherefore be subjective, which consequently renders them less useful. Several tech-nologies and infrastructures supporting rice production developed based on the char-acterization and classification of rice production environments in the past have provento be unsuitable or unsustainable when they are viewed environmentally and socio-economically. Many farmers in several countries still plant many traditional rice vari-eties regardless of the availability of improved rice varieties and extension efforts.Similarly, many large irrigated rice schemes built in Africa during the 1970s and


1980s have not proven to be economically profitable. Insect control techniques basedsolely on chemical application were unsustainable. Therefore, the participation of allstakeholders in environmental characterization, in the selection of production sites,and in the development and selection of technologies needs to be encouraged. ManyFAO field projects, including projects on rice research and development, have re-cently employed the participatory approach (with stakeholders and farmers) duringtheir implementation (Nguyen 1998).

The costs and scale of characterization and classificationEnvironmental characterization and classification also need to be evaluated in termsof returns to agricultural production. More detailed characterization would lead tomore precise classification and thus better inputs for assessing investments, but it isalso more costly. As a rule of thumb, the more details of the characterization, thelarger the scale or ratio between the area under characterization and the area in map-ping of AEZ, and the higher the cost involved. Detailed characterization and classifi-cation are relative, depending on the objectives of the exercise. Some general charac-terization at a smaller mapping scale is adequate for certain objectives, whereas de-tailed characterization at a larger mapping scale is needed for other objectives.Andriesse et al (1994) proposed several levels and scales for the agroecological char-acterization of inland valleys in West Africa (Table 7).

For agricultural production at different administrative levels, Koohafkan et al(1998) proposed four scales and levels of analysis in the characterization of lands andwater-use planning and management (Table 8).

Conclusions

The slowdown in the growth of the world’s rice production after 1990 has led to anincreasing call for the renewal of efforts to sustain rice production. Historical evi-dence indicates that efforts to increase the world’s rice production could be strength-ened by the characterization of rice production environments. As in the past, futureenvironmental characterization should provide the basis for the efficient develop-ment of improved rice technologies and new rice area, the effective transfer of tech-nologies, and other profitable investments aimed at supporting sustainable rice pro-duction. Rice production environments have evolved substantially during the past 30years. Factors affecting rice production and changes in these factors therefore need tobe considered in any new characterization and classification. Future activities in thecharacterization and classification of rice production environments also need to con-sider issues such as universality and compatibility, completeness, objectiveness, andthe costs and scales of the exercises.

50 Van Nguu Nguyen

Tabl

e 7

. Le

vels

and

sca

les

for

agro

ecol

ogic

al c

hara

cter

izat

ion

of in

land

val

leys

.

Cha

ract

eriz

atio

n le

vel

Sca

leU

nit

of a

naly

sis

Obj

ectiv

e

Mac

ro le

vel

1:1

,00

0,0

00

Agro

ecol

ogic

alC

hara

cter

izat

ion

of a

groe

colo

gica

l zon

es s

ubdi

vide

d in

to a

groe

colo

gica

l uni

ts o

n th

e ba

sis

of l

and

to 5

,000,0

00

zone

regi

ons

(land

form

and

lith

olog

y).

Rec

onna

issa

nce

1:1

00

,00

0 t

oC

ount

ryC

hara

cter

izat

ion

of a

groe

colo

gica

l sub

units

on

the

basi

s of

pre

cipi

tatio

n, le

ngth

of

hum

id p

erio

d,le

vel

25

0,0

00

land

form

, lith

olog

y, d

rain

age

dens

ity, m

ajor

upl

and

soils

, maj

or la

nd u

se, a

nd p

opul

atio

n de

nsity

.S

elec

tion

of k

ey a

reas

.S

emid

etai

led

leve

l1:2

5,0

00 t

oK

ey a

rea

Cha

ract

eriz

atio

n of

key

val

ley

syst

ems

base

d on

soi

ls a

nd v

alle

y m

orph

olog

y, p

erio

d of

flo

odin

g5

0,0

00

and

shal

low

gro

undw

ater

, si

ze o

f w

ater

shed

s, la

nd-u

se r

atio

(pe

r la

nd s

ettle

men

t an

d at

val

ley

leve

l), c

rops

and

cro

p ro

tatio

n, s

ocio

econ

omic

fac

tors

(m

arke

t, c

redi

t, e

tc.)

, an

d in

fras

truc

ture

.S

elec

tion

of in

land

val

leys

.D

etai

led

leve

l1

:5,0

00

to

Inla

nd v

alle

yC

hara

cter

izat

ion

of in

land

val

leys

on

the

basi

s of

var

iatio

n of

soi

ls a

nd v

alle

y m

orph

olog

y , s

oil

10

,00

0fe

rtili

ty a

nd t

oxic

ity, so

il ph

ysic

s (in

filtr

atio

n, p

erm

eabi

lity)

, flo

odin

g an

d gr

ound

wat

er d

ynam

ics,

farm

ing

syst

ems

and

crop

ping

int

ensi

ty,

inpu

ts-o

utpu

ts,

crop

var

ietie

s, a

nd c

ropp

ing

cale

ndar

.Q

uant

ifica

tion

of c

onst

rain

ts.

Sou

rce:

And

riess

e et

al (

1994

).

Tabl

e 8

. S

cale

s of

land

-use

and

wat

er-u

se p

lann

ing

and

man

agem

ent.

Leve

l of an

alys

isS

cale

/spa

tial

Issu

ere

solu

tion

Fiel

d/pr

oduc

tion

unit

< 1

:5,0

00

Prod

uctiv

e cr

ops

and

anim

als,

con

serv

atio

n of

soi

l an

d w

ater

, hi

gh lev

els

of s

oil fe

rtili

ty,

low

lev

el o

f so

il an

d(s

ite-s

peci

fic)

wat

er p

ollu

tant

s, lo

w le

vel o

f cr

op p

ests

and

ani

mal

dis

ease

s.Fa

rm/v

illag

e1:1

,000 t

oVi

able

cro

p pr

oduc

tion

syst

ems,

foo

d re

quirem

ents

, ec

onom

ic a

nd s

ocia

l ne

eds

satis

fied,

aw

aren

ess

by(lo

cal l

evel

)50,0

00

farm

ers.

Cou

ntry

(na

tiona

l/1:2

5,0

00 t

oJu

dici

ous

deve

lopm

ent

of a

groe

colo

gica

l po

tent

ial

and

use

of i

rrig

atio

n w

ater

res

ourc

es,

drou

ght

and

flood

subn

atio

nal)

2,5

00

,00

0ris

ks, f

ood

prod

uctio

n an

d fo

od s

ecur

ity, c

onse

rvat

ion

of n

atur

al res

ourc

es a

nd b

iodi

vers

ity, l

and

degr

adat

ion,

publ

ic a

war

enes

s.C

ontin

ent/

wor

ld1

:1,0

00

,00

0 t

oLa

nd d

egra

datio

n an

d de

sser

tific

atio

n, c

onse

rvat

ion

of b

iodi

vers

ity, w

ater

bas

in m

anag

emen

t and

wat

er s

harin

g,5

,00

0,0

00

wat

er p

ollu

tion,

pop

ulat

ion

grow

th a

nd f

ood

secu

rity,

clim

ate

chan

ge a

nd a

gric

ultu

ral p

oten

tial,

awar

enes

s of

regi

onal

and

glo

bal i

nstit

utio

ns.

Sou

rce:

Koo

hafk

an e

t al

(1

998

).


ReferencesAlbergel J, Lamachere JM, Lidon B, Mokadem A, Vandriel W. 1993. Mise en valeur agricole

des bas-fonds au Sahel – Typologie, fonctionnement hydrologique, potentialités agricoles.Rapport final d’un projet CORAF-R3S. Ouagadougou, Burkina Faso. 335 p.

Andriesse W, Fresco LO. 1991. Characterization of rice growing environments in West Africa.Agric. Ecosyst. Environ. 33:377-395.

Andriesse W, Fresco LO, van Duivenbooden N, Windmeijer PN. 1994. Multi-scale character-ization of inland valley agro-ecosystems in West Africa. Neth. J. Agric. Sci. 42(2):159-179.

Antoine J. 1994. Linking geographic information systems and FAO’s agro-ecological zonemodels for land resource appraisal. In: World Soil Resources Report 75–AEZ in Asia.Rome (Italy): Food and Agriculture Organization of the United Nations. p 35-52.

Becker L, Diallo R. 1992. Characterization and classification of rice agro-ecosystems in Côted’Ivoire. Bouaké (Côte d’Ivoire): West Africa Rice Development Association. 252 p.

Bowles EJ, Garrity DP. 1988. Development of a comprehensive classification of rice ecosys-tems. Manila (Philippines): International Rice Research Institute.

Cardenas AC, Dilag RT, Pantastico Jr. EB, Haws LD. 1980. An approach to rainfed farming.Laguna (Philippines): Philippine Council for Agriculture and Resources Research andDevelopment. 27 p.

Cassman KG, Pingali PL. 1995. Intensification of irrigated rice systems: learning from the pastto meet future challenges. GeoJournal 35:299-305.

Cassman KG, Olk DC, Dobermann A. 1997. Scientific evidence of yield and productivitydeclines in irrigated rice systems of tropical Asia. IRC Newsl. 46:7-18.

Chandler Jr. RF. 1979. Rice in the tropics: a guide to the development of the national programme.Boulder, Colorado (USA): Westview Press. 256 p.

FAO (Food and Agriculture Organization of the United Nations). 1976. Framework for landevaluation. Rome (Italy): FAO.

FAO (Food and Agriculture Organization of the United Nations). 1978. World Soil ResourcesReport 48/1 – Africa. Rome (Italy): FAO.

FAO (Food and Agriculture Organization of the United Nations). 1980. World Soil ResourcesReport 48/4 – Asia. Rome (Italy): FAO.

FAO (Food and Agriculture Organization of the United Nations). 1981. World Soil ResourcesReport 48/3 – Central and South America. Rome (Italy): FAO.

FAO (Food and Agriculture Organization of the United Nations). 1994. World Soil ResourcesReport 75 – AEZ in Asia. Rome (Italy): FAO.

FAO (Food and Agriculture Organization of the United Nations). 1997. Trends of yield andproductivity of modern rice in irrigated rice systems in Asia. IRC Newsl. 46:19-28.

Huke RE. 1982a. Map of South and Southeast Asia. Manila (Philippines): International RiceResearch Institute.

Huke RE. 1982b. Rice area by type of culture: South, Southeast, and East Asia. Manila (Philip-pines): International Rice Research Institute.

IRRI (International Rice Research Institute). 1974. An agro-climatic classification for evaluat-ing cropping systems potentials in Southeast Asian rice-growing regions. Los Baños(Philippines): IRRI. 10 p.

IRRI (International Rice Research Institute). 1984. International terminology of rice growingenvironments. Manila (Philippines): IRRI.

52 Van Nguu Nguyen

Jamin JY, Andriesse W. 1993. Discussion synthesis. In: Proc. 1st Annual Workshop of InlandValley Consortium, 8-10 Oct 1993. Bouaké (Côte d’Ivoire): West Africa Rice Develop-ment Association. p 8-16.

Karim Z. 1994. Cropping systems based fertilizer recommendations by agro-ecological zonesin Bangladesh. In: World Soil Resources Report 75 – AEZ in Asia. Rome (Italy): Foodand Agriculture Organization of the United Nations. p 53-77.

Khush GS, Bennett J, Datta SK, Brar DS, Li Z. 1999. Advances in rice genetics and biotechnol-ogy. In: Proceedings of the 19th Session of the IRC, 7-9 Sep 1998, Cairo, Egypt. Rome(Italy): Food and Agriculture Organization of the United Nations. p 64-76.

Koohafhan P, Nachtengale F, Antoine J. 1998. Use of agro-ecological zones and resourcesmanagement domains for sustainable management of African wetlands. In: Proceedingsof a sub-regional consultation workshop on Wetland Characterization and Classificationfor Sustainable Agricultural Development, 3-6 Dec 1997, Harare, Zimbabwe. p 107-132.

Legoupil JC, Bidon B. 1993. What level of water control for inland valley intensification inWest Africa? In: Proceedings of the first Workshop of the IVC held from 8-10 Jun 1993.Bouaké (Côte d’Ivoire): West Africa Rice Development Association. p 45-60.

Nguyen VN. 1998. Factors affecting wetland rice production and the classification of wetlandfor agricultural development. In: Proceedings of a sub-regional consultation workshopon Wetland Characterization and Classification for Sustainable Agricultural Develop-ment, 3-6 Dec 1997, Harare, Zimbabwe. p 175-90.

Oldeman LR. 1974. An agro-climatic classification for evaluation of cropping systems in South-east Asia. Paper presented at FAO/UNDP international expert consultation on the use ofimproved technology for food production in rainfed areas of tropical Asia held atHyderabad, India, 24-30 Nov; Khon Kaen, Thailand, 1-7 Dec; and Kuala Lumpur,Malaysia, 8-13 Dec 1974.

Oldeman LR, Frere M. 1982. A study of the agro-climatology of the humid tropics of SoutheastAsia. Rome (Italy): FAO.

Pardey PG, Wood SR. 1991. Targetting research by agricultural environments. Chapter 31 in:Anderson JR, editor. Agricultural technology policy: issues for the international com-munity. Wallingford (UK): CAB International.

Penning de Vries EWT, Jansen DM, ten Berge HFM, Bakema A. 1989. Simulation ofagroecological processes of growth in several annual crops. Manila (Philippines): Inter-national Rice Research Institute.

Pingali PL, Rosegrant MW. 1994. Confronting the environmental consequences of the greenrevolution. In: Proceedings of the Eighteenth Session of the International Rice Commis-sion, 5-9 Sep 1994, Rome, Italy. p 59-69.

Pingali PL, Hossain M, Gerpacio RV. 1997. Asian rice bowls: the returning crisis? Wallingford(UK): CAB International.

Setter TL, Ingram KT, Tuong TP. 1995. Environmental characterization requirement for strate-gic research in rice grown under adverse conditions of drought, flooding, or salinity. In:Ingram KT, editor. Rainfed lowland rice: agricultural research for high-risk environ-ments. Manila (Philippines): International Rice Research Institute. p 3-18.

Ton That T. 1982. Potentialities and constraints of rainfed lowland rice development in tropicalAfrica. IRC Newsl. 31(2):1-6.

Tran DV, Nguyen VN. 1998. Global hybrid rice: progress, issues and challenges. IRC Newsl.47:16-28.


Uehara G, Tsuji GY. 1991. Progress in crop modelling in IBSNAT project. In: Muchow RC,Bellamy JA, editors. Climate risk in crop production: models and management for thesemi-arid tropics and subtropics. Wallingford (UK): CAB International. p 143-156.

Windmeijjer PN, Andriesse W. 1993. Inland valleys in West Africa: an agro-ecological charac-terization of rice-growing environments. The Netherlands: International Institute forLand Reclamation and Improvement.

Wood SR, Pardey PG. 1993. Agro-ecological dimensions of evaluating and prioritizing re-search from a regional perspective: Latin America and the Caribbean. The Hague (TheNetherlands): International Service of National Agricultural Research.

NotesAuthor’s address: Agricultural Officer, Crop and Grassland Service, Plant Production and Pro-

tection Division, Agriculture Department, Food and Agriculture Organization of theUnited Nations, Vialle delle Terme di Caracalla, 00100 Rome, Italy.

Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-acterizing and understanding rainfed environments. Proceedings of the InternationalWorkshop on Characterizing and Understanding Rainfed Environments, 5-9 Dec. 1999,Bali, Indonesia. Los Baños (Philippines): International Rice Research Institute.488 p.

Effect of climate, agrohydrology, and management . . . 55

Tools and methodologiesfor biophysical characterization

56 Boling et al


The typical rainfed cropping system in Central Java includes a dry-seededrice crop grown from November to February (gogorancah), followed by a trans-planted rice crop from March to June (walik jerami). Earlier studies showedthat the yield of the walik jerami crop was lower and less stable than that ofthe gogorancah crop. This study assessed the climatic and agrohydrologic(groundwater depth) constraints to rice production and explored manage-ment strategies to increase the yield and yield stability of the double-ricecropping system using the crop growth simulation model ORYZA. The modelwas validated with data of field experiments in 1995-96 in Jakenan. Long-term simulation of potential and rainfed rice yield of cultivar IR64 was carriedout on a 15-d planting interval for the period 1977-98. Three water tabledepth scenarios (medium, shallow, and deep), which were derived from 1995to 1999 measurements, were used. The average simulated potential yield ofwalik jerami rice (about 7 t ha–1) was higher than that of gogorancah rice(about 6 t ha–1), indicating that radiation and temperature are not the deter-minants of the observed relatively low yields of walik jerami rice. Simulatedyields of rainfed rice sown with a shallow water table depth in mid-November-February equaled the potential yield. Rainfed rice yield was reduced by 45%in the medium water table scenario and by 70% in the deep water tablescenario. With medium water table depths, simulated rainfed yields of walikjerami crops declined sharply if planted later than early March. Supplementalirrigation increased the yields of rainfed walik jerami crops. The combinedyields of gogorancah and walik jerami rice could be increased by using shorter-duration varieties. The results highlight the critical planting dates of the walikjerami crop, and indicate the potentials of using on-farm reservoirs and ofgrowing shorter-duration varieties to increase the combined yields and yieldstability in the area.

Effect of climate, agrohydrology,and management on rainfedrice production in Central Java,Indonesia: a modeling approachA. Boling, T.P. Tuong, B.A.M. Bouman, M.V.R. Murty, and S.Y. Jatmiko

58 Boling et al

Rainfed lowland rice in Central Java covers about 293,600 ha, which is equivalent to30% of the rice area (Amien and Las, this volume). In this area, farmers practice ahigh degree of crop intensification. The common cropping pattern includes two rainfedrice crops (see Fig. 1A). At the onset of the rainy season, a dry-seeded rice crop isgrown, called gogorancah. Immediately after the harvest of gogorancah, a second,transplanted rice crop is grown under minimum tillage in submerged conditions. Thisparticular cultivation of the second rice crop is called walik jerami. If residual soilmoisture is still adequate, or where irrigation from shallow groundwater or from on-farm water reservoirs is possible, farmers may afterwards grow an upland crop in thedry season (called palawija).

Earlier studies showed that the average yield of gogorancah rice is 4–6 t ha–1,whereas walik jerami yields are only 1.5–3.0 t ha–1 (Fagi 1995, Mamaril et al 1995,Wihardjaka et al 1998). According to Fagi (1995), Mamaril et al (1995), and Wihardjakaet al (1998), the relatively low and unstable yield of the walik jerami crop is attributed

Fig. 1.Cropping system (A), daily solar radiation (B), and 10-d rainfall(C) in Jakenan, Central Java, Indonesia.

Uplandcrop (palawija)

Transplanted rice(walik jerami)

Dry-seeded rice(gogorancah)

ACropping system

22

20

18

16

14

12

10

8

Solar radiation (MJ m–2 d–1)

B

Rainfall (mm 10 d–1)

C160

140

120

100

80

60

40

20

0O N D J F M A M J J A S O

Exceedance probability (P)P = 0.20P = 0.50P = 0.80

Month


to water deficit because the crop’s reproductive stage coincides with the recession ofrainfall in the area. Wihardjaka et al (1998) further attributed the higher yield ofgogorancah crops to more favorable temperature and radiation regimes. To date, thequantitative effect of temperature, radiation, and rainfall on yield of both thegogorancah and the walik jerami crop in Central Java has yet to be determined.

The rainfed lowland areas in Central Java are characterized by an undulatinglandscape. In these areas, the water table depth probably varies across the toposequence.Shallow water tables at the bottom of the toposequence may contribute significantlyto the water requirements of the rice crop via capillary rise. On the other hand, deepwater tables at the top of the toposequence may have a negligible contribution to thecrop’s water requirements. Therefore, the position of a rice field in the landscape mayhave a pronounced effect on rice yields. Pests and diseases may also depress riceyields.

Several management strategies that alleviate the effect of drought in the areacan be explored. Planting dates of both the gogorancah and the walik jerami cropscan be adapted so that the combined yields are maximized by avoiding the plantingdates that expose the crops to adverse conditions. The recent introduction of on-farmwater reservoirs (Syamsiah et al 1994) offers scope for supplemental irrigation, espe-cially to alleviate drought during the reproductive stage of the walik jerami crop atthe recession of the rainy season. The use of shorter-duration varieties could mini-mize the exposure to drought of the walik jerami crop during its reproductive stagesince shorter-duration varieties advance the flowering date compared with the com-monly used varieties. The relative increase in rice yield because of drought escape,however, may be offset by the lower yield potentials of shorter-duration crops. Theimpact of the above drought alleviation strategies remains to be quantified in CentralJava.

Simulation models have been used to evaluate rainfed rice ecosystems and toexplore management strategies for constraint alleviation (e.g., Wopereis et al 1995,Jongdee et al 1997). Our study uses the ecophysiological crop growth model ORYZAto (1) assess the climatic and agrohydrologic (groundwater depth) constraints to rainfedrice production in Central Java, and (2) explore the efficacy of management strategiesaimed at increasing yield and yield stability of the double-rice cropping system.

Materials and methods

Site descriptionThe study was carried out at the Jakenan Experiment Station (6°45′S, 111°10′E, 7 mabove sea level). The landscape is undulating. The soil is alluvial, with 20-cm light-textured surface soil (44% sand, 46% silt, 11% clay) and a clayey 21–40-cm subsur-face layer (35% sand, 34% silt, 31% clay). The top 20-cm soil layer has a dry bulkdensity of 1.46 g cm–3, low organic carbon (0.41%), relatively low CEC (3.48 meq100 g–1), and low exchangeable bases (0.03 K, 2.80 Ca, 0.22 meq Mg 100 g–1). Theexperimental site represents a rainfed lowland area that covers about 150,000 ha inCentral Java (Mamaril et al 1995).

60 Boling et al

According to Oldeman (1975), the climate of the site can be classified as D3.Maximum temperature is 31.7 ± 0.1 °C and minimum temperature is 23.5 ± 0.1 °Cthroughout the year. Solar radiation is low from December to February and high fromAugust to October (Fig. 1B). The rainy season usually starts in October, peaks inJanuary, and ends in May or June (Fig. 1C). The average annual rainfall (1953-98) is1,540 mm, with 1,050 mm falling in the 5-mo period from November to March. Thearea has a rather long (≥5 mo) growing season (monthly rainfall exceeds evapotrans-piration, Garrity et al 1986).

ExperimentsField experiments using rice variety IR64 were conducted in April-July 1995 andMarch-July 1996 at the Jakenan Experiment Station for model calibration and evalu-ation. In 1995, rice was transplanted on 8 April and treatments were laid out in a split-plot design with four replications using two water treatments in the main plot andthree tillage treatments in the subplots. The water treatments were full irrigation andno irrigation (rainfed). The main plots were lined with polyethylene sheets up to 40-cm depth to minimize subsurface lateral water flow. The tillage treatments were nor-mal tillage (hoed once to 10-cm depth), deep tillage (hoed twice to 30-cm depth), anddeep tillage with puddling (puddled after hoeing). In 1996, the experiment includedtwo water treatments (fully irrigated and rainfed) in the main plot, two tillage (normaland deep) treatments in the subplots, and two transplanting dates (10 March and 8April) in the sub-subplots. The staggered transplanting dates were used to expose therice to different levels of drought stress. A basal fertilizer equivalent to 30 N, 22.5 P,30 K, and 20 kg S ha–1 was applied a day before planting. Additional fertilizer equiva-lent to 60 N and 30 kg K ha–1 was applied at maximum tillering, 30 kg N ha–1 atpanicle initiation, and 30 kg K ha–1 at flowering. Hand weeding and pesticide spray-ing were used to minimize pest damage.

Daily rainfall, solar radiation, maximum and minimum temperature, relativehumidity, and wind speed were measured at the Jakenan weather station. We mea-sured biomass and its partitioning (leaves, culm and sheath, and panicle) at 30, 45,and 66 d after transplanting (DAT), at flowering, and at physiological maturity. Riceyield was determined from an 8-m2 sampling area at harvest.

During the periods when there was no water standing in the fields, soil waterpotentials in the rainfed plots were measured daily using tensiometers installed at 5-,10-, 20-, and 30-cm depth. The soil water content at the same depths was measuredwhen soil water potentials were beyond the air entry values (approximately 60–80 kPa) of the tensiometers.

The groundwater table for the experiments was measured daily in four of therainfed plots using 2.5-cm-diameter, 150-cm-long PVC tubes perforated with 3-mm-diameter holes along a 100-cm length from the bottom. In addition, we used the ground-water table measurements during the period April-June 1997, December 1997-June1998, and November 1998-February 1999 to characterize the water table fluctuation.Since we measured the groundwater table depths only during the crop growth period,the measuring period was not continuous. The number of measurements thus differed


for different periods of the year: four sets of data for April-May, three sets for June,two sets for December-January, and one set for October, November, February, andMarch. No data were measured during the July-October peak of the dry season. Dur-ing this period, the water table was assumed to be 150 cm below the soil surface.

To quantify the fluctuation of the groundwater table depths, we constructed theshallow, medium, and deep water table scenarios as follows:

1. For periods with more than one measurement, the means of the measuredvalues were used to represent the medium water table depths, the mean +standard error of measured values to represent the shallow depths, and mean– standard error for the deep water table. The medium water table scenariowas presented by the line connecting the calculated medium depths. Theenveloping line connecting the mean + standard error represented the shal-low water table scenario, and the line connecting the mean – standard errorthe deep water scenario.

2. For periods with only one measurement, the measured values were used forthe medium water table scenario. The line that connects the shallow watertable depths of the adjacent periods (with more than one measurement) wasused to represent the shallow water table scenario and the line that connectsthe deep water table represents the deep water table scenarios.

3. For the period at the peak of the dry season (i.e., July-October), the watertable was assumed to be at 150 cm below the soil surface in the three watertable scenarios.

The shallow, medium, and deep water table scenarios were used as inputs to themodel simulating rainfed rice production.

The ORYZA modelORYZA is an updated and integrated version of the ecophysiological models ORYZA1(Kropff et al 1994) and ORYZA_W (Wopereis et al 1996). ORYZA simulates cropgrowth and development of lowland rice in potential and water-limited productionsituations. Under potential situations, water and nutrients are in ample supply andgrowth rates are determined by weather conditions only (radiation and temperature).Under water-limited production, growth is limited by water shortage in at least part ofthe growing period, but nutrients are still considered to be in ample supply. In bothproduction situations, the crop is supposed to be well protected against pests, dis-eases, and weeds. In our study, water-limited production is synonymous with rainfedproduction.

ORYZA consists of separate modules to calculate growth and development ofthe crop, evapotranspiration, effects of drought on growth and development, and thewater balance of puddled and nonpuddled soils. The crop module is a photosynthesis-driven model. On each day, the solar radiation profile in the canopy is calculated onthe basis of incident radiation, leaf area index, and the vertical distribution of leavesin the canopy. The daily canopy assimilation rate is calculated by integrating thecalculated photosynthesis of single leaves over the height of the canopy and over theday. After subtracting respiration requirements and accounting for losses from the

62 Boling et al

conversion of carbohydrates into structural dry matter, the net daily growth rate isobtained. The dry matter produced is partitioned among the various plant organs ac-cording to the stage of development of the crop, which is tracked as a function ofambient mean daily air temperature.

Daily evapotranspiration rates are calculated using modified Penman equations(Van Kraalingen and Stol 1997). Effects of drought (defined as the condition whensoil water contents are lower than saturation) on crop growth and development in-clude leaf rolling, reduced leaf growth rate, accelerated leaf senescence, reduced evapo-transpiration and photosynthesis, reduced development rate, reduced sink size, andreduced spikelet fertility. The drought stress response functions were derived frompot experiments (Wopereis 1993, Wopereis et al 1996; unpublished experiments bythe authors) and the literature (Turner et al 1986). The soil water balance, calledPADDY, is a universal multiple-layer model that can be used for both puddled andnonpuddled conditions, for freely draining soils, and for soils with impeded drainage(Wopereis et al 1996). Capillary rise from groundwater into the rooting zone of thesoil profile is taken into account.

ORYZA requires input data on crop characteristics, soil properties, manage-ment, daily water table depth, and daily weather. ORYZA was extensively calibratedand evaluated for potential production situations (Kropff et al 1994, Matthews et al1994, 1995). The model was validated for rainfed production at the IRRI farm in LosBaños, Laguna, Philippines (Wopereis 1993, Wopereis et al 1996).

Model simulationsFour series of simulation runs were made. The first series consists of model calibra-tion and performance evaluation. ORYZA was parameterized and evaluated using the1995-96 experimental data and the observed daily weather data. For the crop, stan-dard physiological characteristics for IR64 were used (IRRI, unpublished data set).The irrigated treatments of both 1995 and 1996—supposedly representative for po-tential production situations—were used to derive the required empirical parametersof development rate, assimilate partitioning, and leaf death rate. The rainfed treat-ments were used for evaluation (independent data set). For all simulations, measuredsoil physical properties and water table depths were used as inputs in ORYZA’s soilwater balance.

The second series of simulations concerned the exploration of long-term riceyield under potential and rainfed conditions. Potential and rainfed yields were calcu-lated for different planting dates at 15-d intervals using historical weather data of1977-98. To keep the simulation as much as possible in line with farmers’ practices,simulation from January to May was performed for transplanted rice, and for the restof the year for direct-seeded rice. The evaluation of potential rice yield sheds light onthe (interactive) contribution of solar radiation and temperature to yield formation inthe gogorancah and walik jerami crops. The simulation of rainfed rice yields quanti-fies the effect of drought on yield reduction and allows us to quantify critical dates forplanting of the walik jerami crop in relation to the recession of the rains at the end ofthe rainy season. The rainfed rice yields were calculated for three groundwater table


depth scenarios (shallow, medium, and deep) to quantify the effect of groundwatercontribution on yield formation. These scenarios were derived from the groundwaterdepth measurements in 1995-99 (see “Experiments”).

The third simulation series focused on the sensitivity of the gogorancah crop toinitial soil moisture at sowing. Transplanting of walik jerami rice is commonly donein saturated soil at the peak of the rainy season. Therefore, model simulations withORYZA can safely begin with saturated soil conditions. On the other hand, the seed-ing of the gogorancah crop starts at the beginning of the rainy season when the initialsoil moisture varies across years depending on rainfall distribution. The initial soilmoisture status during the establishment of the gogorancah crop may affect the veg-etative growth and yield of rice crops grown in the area, and therefore warrants asensitivity analysis with ORYZA. Four scenarios, that is, initial soil moisture status atsaturation, field capacity, –100 kPa soil water potential, and at permanent wiltingpoint, were simulated.

The fourth simulation set explored the effect of supplementary irrigation and ofthe use of relatively short-duration varieties on rice yield. Three supplementary irri-gation scenarios were constructed: (1) irrigation water was applied whenever the top-soil water content fell to field capacity (= 0.34 cm3 cm–3, scenario I1), (2) a dailyamount of 7.5 mm water was applied from panicle initiation to crop maturity (sce-nario I2), and (3) a daily amount of 3.3 mm water was applied from panicle initiationto crop maturity (scenario I3). To study the effect of crop duration on rice yield, simu-lations were performed, besides for IR64 (V1), for two hypothetical varieties with 5-d (V2) and 10-d (V3) reduction in growth duration compared with IR64. The reduc-tion in growth duration was accomplished by diminishing the parameterized valuefor the development rate of IR64. In the supplementary irrigation scenarios, irrigatedyield was simulated (i.e., rainfed plus supplementary irrigation) and, in the crop dura-tion scenarios, both potential and rainfed situations were simulated, with 15-d-inter-val planting dates and using historical weather data of 1977-98.

Results and discussion

Model evaluationFigure 2 gives simulated and measured total dry canopy biomass in 1996. The simu-lated potential biomass falls mostly within the standard error range of measured bio-mass in the highest yielding irrigated plot (deep tillage treatment, supposedly the besttreatment approaching potential production situations). It is to be expected that simu-lated potential biomass values are higher than measured biomass values in the irri-gated treatments since it is extremely difficult to realize potential growth conditionsin field experiments (e.g., Kropff et al 1994). Some yield reduction might have oc-curred because of nutrient limitation, pests, and diseases. Even so, simulated rainfedbiomass values were generally somewhat higher than measured biomass values in thetwo rainfed treatments with different sowing dates.

ORYZA mostly performed about the same in the simulation of yield as in thesimulation of total biomass (data not shown). Only under severe drought stress occur-

64 Boling et al

ring during the reproductive phase did ORYZA overestimate grain yield by a factor of4 to 10 (measured values were 200–500 kg ha–1 while simulated values were 2,000 kgha–1). Whereas ORYZA partitions all assimilates produced in the reproductive phaseto the storage organs, it was observed that, in reality, when the stress was relieved, thecrop produced new tillers, indicating that not all assimilates went to the grains asmodeled.

Though a statistical analysis of model performance still needs to be made, thepreliminary results indicate that ORYZA performed sufficiently well for the purposesof our study.

Water table fluctuationFigure 3 presents the shallow, medium, and deep water table scenarios constructedfrom measurements in six cropping seasons of 1995-99. Large standard errors (inperiods with more than one measured value) indicated that the water table depth var-ied widely in different years. The large standard errors, however, may be attributed tothe small number of available data. The mean water table depth fluctuated from 10cm below the soil surface in December to around 30–60 cm in January and in April.The mean water table started to drop off from the middle of May to 70–100 cm inJune. This conforms to the declining rainfall in these periods and suggests that, start-ing from June, the contribution of the capillary water from the groundwater to theroot zone decreased substantially. The water table depth in February-March appearedvery deep and might not adequately reflect the “medium” value because we had onlyone set of data that were measured in 1998, a year with particularly low rainfall.

Fig. 2.Simulated and measured aboveground dry matter (mean ± SE)during the 1996 walik jerami season in Jakenan. The solid line is thesimulated potential production, which may be compared with the mea-surements of the best irrigated treatment. The broken and dottedlines are the simulated rainfed productions using early and late sow-ing. The measurements of the rainfed treatments include two tillagetreatments.

16,000

12,000

8,000

4,000

0

Irrigated, measuredEarly rainfed, measuredLate rainfed, measuredPotential, simulatedEarly rainfed, simulatedLate rainfed, simulated

Dry matter (kg ha–1)

F M A M J J A

Month


Yield explorationPotential yield. The simulated, long-term average potential rice yield is shown inFigure 4 as a function of date of sowing. The potential yields ranged from 6 to8 t ha–1, and were within the yield range of well-managed irrigated rice as observed inBogor, West Java, and in Genteng, East Java, by Makarim and Las (1993). The stan-dard errors of the average values were low, indicating little variation in potential yieldacross years. Rice sown in the typical gogorancah period (November-December) hada lower potential yield (on average 6 t ha–1) than rice planted in the typical walikjerami period (February-March, average yield = 7 t ha–1). This lower potential yieldin the gogorancah period was caused mainly by the low radiation during the crop’sreproductive stage (Fig. 1B). The variation in potential yield for crops sown in No-vember-December, however, was relatively larger than for crops sown in February-March.

Rainfed yield. The simulated, long-term average rainfed rice yield is shown inFigure 4 as a function of date of sowing for situations with a shallow, medium, anddeep groundwater table. The yield of rainfed rice sown from mid-November to theend of March differed significantly among the three water table depth scenarios. Dur-ing this period, rainfed yields reached the potential yield level with the shallow watertable, but decreased with deeper water table depths. For crops sown at the end of therainy season and in the dry season (roughly April-November), the yield was very lowand yield differences among the water table scenarios disappeared. The low yields inthis season are attributed to inadequate water supply from rainfall and from the ground-

Fig. 3.Shallow, medium, and deep water table scenarios used inORYZA simulations to explore rainfed rice yields in Jakenan. Themethod of deriving the scenarios from 1995-98 measured data isdescribed in the text. Solid triangles indicate the means of the mea-sured data (or the actual data when there was only one water tablevalue measured). Vertical bars indicate standard errors of the means(when there was more than one water table value).

50

0

–50

–100

–150

Water table depth (cm)

O N D J F M A M J J A S O

Month

Shallow water tableMedium water tableDeep water table

66 Boling et al

water. Considering the overestimation of simulated yield under severe drought condi-tions by ORYZA (see “Model evaluation”), we can in fact assume that simulatedrainfed yields in the dry season were virtually zero.

In the shallow water table scenario, the yield of crops sown in the typicalgogorancah period (November-December) was significantly lower than that of cropssown in the typical walik jerami period (February-March) (similar to potential yielddifferences; see above). However, in the deep water table scenario, the opposite istrue: rainfed rice yields are higher in the typical gogorancah period than in the typicalwalik jerami period. This latter is consistent with field studies reported by Fagi (1995),Mamaril et al (1995), and Wihardjaka et al (1998).

Sensitivity analysisFigure 5 shows the simulated rainfed yield of gogorancah rice sown in October-December with medium water table depths at four levels of initial soil moisture con-tent. In general, rice sown in soil with a high initial moisture content had a high yield.The observed yield differences among initial soil moisture statuses decreased as theplanting dates moved from October to December. Simulated yield of rainfed ricecrops sown at a given soil moisture differed among seeding dates. Yields of rainfedrice sown in a low soil moisture status in October-early November were significantlylower than those of crops sown from mid-November onward. The yield difference ofrainfed rice for crops sown before and after mid-November, however, was consider-ably lower in soils with high initial moisture contents. For crops sown from mid-

Fig. 4.Simulated potential and rainfed rice yield (mean ± SE) inJakenan, Central Java, Indonesia.

Simulation fordry-seeded rice

Simulation fortransplanted rice

Simulation fordry-seeded rice

O N D J F M A M J J A S O

Day of seeding by month

12,000

10,000

8,000

6,000

4,000

2,000

0

Simulated yield (kg ha–1)

Rainfed, shallow water tableRainfed, medium water tableRainfed, deepwater tablePotential


November onward, the relatively high yields obtained regardless of initial soil mois-ture status indicate that rice yield is stable if rice is sown after early November.

Management options to increase yieldEstablishment date. Figure 4 shows that the three water table scenarios caused theyield of rainfed crops to start to decline at different seeding dates. Yields in the shal-low water table scenario started to decline in mid-March, whereas yields in the deepwater table scenario started to decline in mid-January (Fig. 4). The yield decline wasopposite to the yield-increasing effect of increasing radiation levels toward the reces-sion of the rainy season (Fig. 1B) and was attributed to water stress because of thedecreasing rainfall during the reproductive stage of the crops (Fig. 1C).

The observed decline in simulated yield of rainfed crops planted toward therecession of the rainy season (Fig. 4) highlighted the importance of planting dates ofthe walik jerami crop. The planting date of the walik jerami crop, however, dependson the harvest date of the preceding gogorancah crop. A late onset of the rains maydelay the seeding of the gogorancah crop, which consequently delays the establish-ment of the walik jerami crop. If the delay is beyond a critical transplanting date, thefarmer may choose whether to forgo the walik jerami crop or risk having a low walikjerami yield. From the simulation results, the critical sowing date of the walik jeramicrop with a shallow water table was mid-March, with a medium water table mid-February, and with a deep water table mid-January.

Supplementary irrigation. The simulated rice yield in the three supplementaryirrigation scenarios with medium water table depth and initial soil water potential of–100 kPa is presented in Figure 6A. For crops sown in the typical walik jerami period

Fig. 5.Simulated yield (mean ± SE) of dry-seeded rice with a mediumwater table depth as a function of sowing date at four different initialsoil moisture contents in Jakenan.

6,000

5,000

4,000

3,000

2,000

1,000

0

Day of seeding by month

S O N D J F

Simulated yield (kg ha–1)

SaturationField capacitySoil water potential = –100 kPaPermanent wilting point

Initial water status:

68 Boling et al

(February-March), simulated yields in irrigation scenarios I1, I2, and I3 were on aver-age about 2.1–3.2 t ha–1, 1.5–3.1 t ha–1, and 0.5–1.2 t ha–1 higher, respectively, thanthe simulated yields under purely rainfed conditions. Figure 6B shows the averagetotal amount of irrigation water applied in the three irrigation scenarios. The irriga-tion water for crops sown in the walik jerami period in scenario I1 varied from 347mm with sowing in mid-February to 555 mm with sowing in mid-April. On the otherhand, the irrigation water for crops sown from mid-February to mid-April in sce-narios I2 and I3 is fixed at 345 mm and 153 mm, respectively.

The increase in rice yields for every m3 of irrigation water applied in the threeirrigation scenarios is presented in Figure 7. The maximum yield increases in thethree scenarios occurred for crops sown in early March. For crops sown from mid-

Fig. 6.Simulated yield of rainfed and irrigated rice (A) andamount of irrigation water in three irrigation scenarios withmedium water table depth (B) as a function of sowing date.I1 = irrigation when moisture of the topsoil falls below fieldcapacity (0.34 cm3 cm–3), I2 = daily irrigation with 7.5 mmfrom panicle initiation (PI) to maturity (M), I3 = daily irriga-tion with 3.3 mm from PI to M.

Perception, understanding, and mapping of soil variability . . . 75

Previous studies that assessed the soil resources in northeast Thailand cre-ate the impression that soils are universally infertile because of their lighttexture and low inherent nutrient contents. In reality, variations within micro-catchments are sufficient to influence land productivity factors over shortdistances. This chapter describes a study carried out to examine soil variabil-ity in the rainfed lowlands of northeast Thailand and to develop quantitativemethods of spatial prediction that provide useful soil resource informationfor agronomic management. Our methodology is based on geostatisticalmapping, using soil data collected in soil surveys supplemented with lesscostly auxiliary information. The auxiliary information included knowledge oflocal farmers and soil experts about soil-landscape relations (providing a 5-category classification referred to in this study as the updated farmers’ fieldclassification, or UFFC). A soil sampling scheme was devised equivalent tothat employed for producing soil maps at a mapping scale of 1:50,000 to1:100,000. Hence, application of the proposed method in areas that havebeen mapped at these scales will only require a reanalysis of existing dataand collection of complementary data at lower costs.

Soils of loamy sand or loam texture were found in many parts of thestudy region, indicating that the soil texture of these soils is not altogetherunsuitable for rice cultivation. Conventional statistical analysis of the soilsurvey data reveals very high spatial variability that cannot be ignored. Manytopsoil and subsoil properties related to nutrient availability (Bray-II P, cationexchange capacity, exchangeable bases) had large coefficients of variation(CVs), including those properties that are considered relatively stable, suchas organic matter and clay content. Further statistical analysis shows thatthis soil map accounts for less than only 8% of the variance in measured soilproperties, not enough to provide agronomically important information. Therewas no discernible distinction in soil fertility characteristics among the mappedsoil types. The UFFC accounted for higher proportions of the total variance(ranging from 0% to 43% for both soil depths), and indicated that soil produc-tivity declines in the order alluvial fields > fields in low topographic position >fields in medium and high topographic position > fields with upper, deepsandy horizons.

Perception, understanding,and mapping of soil variabilityin the rainfed lowlandsof northeast ThailandT. Oberthür and S.P. Kam

76 Oberthür and Kam

Geostatistical indicator approaches that can either use or ignore auxil-iary information were adopted for modeling of soil heterogeneity to enableestimation and mapping of mean, median, conditional variance, and condi-tional CV, and also of the 0.2- and 0.8-quantiles of the conditional cumulativedistribution function. Quantile maps are very useful for mapping soil proper-ties tailored to specific land management questions. Maps produced withoutauxiliary information show distinct spatial distribution patterns of averageclay, silt, and sand contents in relation to the regional physiography. Mapsgenerated with auxiliary information reveal more spatial detail; texture changesgradually and follows the local topography and drainage patterns. These mapsassign much land (25%) to a soil texture class that is suitable for rice produc-tion (loam and heavier).

In conclusion, our results depart from the general belief that soils ofthe Korat Plateau are universally coarse-textured and infertile. The relianceon soil maps has contributed to the long-held views about soil texture innortheast Thailand. Our maps strongly support an alternative hypothesis ofsoil genesis in the Korat Plateau that combines colluvio-alluvial erosion pro-cesses over short distances and in situ soil development with long-rangeQuaternary alluvial sedimentation.

Understanding the heterogeneity of the environment is a cornerstone to increasingand sustaining rice productivity in the rainfed lowlands (Zeigler and Puckridge 1995).Modern techniques for generating and analyzing geographic information (geographicinformation systems, remote sensing, and global positioning systems) used in con-junction with systems approaches can help improve environmental characterization.So far, however, data-intensive techniques have been attempted only in irrigated sys-tems at the field scale in the rice-growing areas of Asia (Dobermann and White 1999).Reliable quantitative data are often lacking beyond specific field locations in rainfedlowland rice areas of Southeast Asia, and this shortcoming impedes a broad-basedassessment of land resources as a prerequisite for agronomic management decisions.This is particularly true for soil information.

Over the past 20 years, land evaluation has benefited from the introduction ofgeographic information systems (GIS), which enable spatial modeling of informa-tion. GIS explicitly provide the spatial dimension for land resource assessment andthe means for integrating data of different subject matters and from disparate sources.Conventional (GIS-based) approaches for deriving soil information at regional andsubregional scales (1:20,000 to 1:100,000), however, suffer from the shortcoming oftraditional sources of soil information, principally soil taxonomic maps. Because oftheir low spatial resolution, focus on pedological characterization, lack of quantita-tive data, large between-farm variability in crop management, and dynamic changesin many soil nutrients, existing soil maps in the developing countries of Asia often donot provide sufficient information for agronomic purposes (Oberthür et al 1996). Spe-cific spatial assessment and monitoring of soil-related constraints to nutrient uptake


and rice yield at regional or more detailed scales are needed to identify problem areasas a basis for making strategic agronomic decisions. This chapter discusses this prob-lem and suggests solutions to improve the availability of soil data and characterizethe associated uncertainty whereby GIS tools are used in conjunction with relatedtechniques such as geostatistics.

Perception of soil variability

Previous studies that assessed the soil resources in northeast Thailand created theimpression that soils are universally infertile because of their light texture and lowinherent nutrient contents (Ragland and Boonpuckdee 1988). These generalizationsmight be grounded in the fact that the finite elements of the assessment were soilseries or soil groups; see, for instance, Suddhiprakarn and Kheoruenromne (1987) orPatcharapreecha (1988). In reality, however, variations within micro-catchments aresufficient to influence land productivity factors over short distances. This short-rangevariation of soil fertility indicators within the landscape has long been recognized asimportant in studies on land resource assessment (Craig and Pisone 1988, Grandstaff1988) but is often ignored or smoothed out in conventional soil maps. Acceptance ofMoormann’s soil development theory (Moormann et al 1964), which implied gradualchanges in soil properties over long distances, has contributed to this practice.

Land formation and soil genesis in northeast Thailand have been the subject ofcontroversial discussions. Pendleton and Montrakun (1960) favored an in situ soildevelopment of particular soil types at certain topographic positions. Moormann et al(1964) refuted the in situ soil genesis hypothesis and linked soil formation to vastPleistocene sediments deposited by the Mekong River and its tributaries. Accordingto Moormann et al (1964), deposition of sediments and soil formation occurred infour distinct phases, resulting in the high, middle, and lower terraces and the presentfloodplain. Recently, however, several workers have independently questioned thistheory. Paiboon et al (1985), Tamura (1986), Mitsuchi et al (1989), and Miura (1990)all support a combination of in situ soil development, mainly through lessivage, andsoil development by colluvial processes over short distances. The topography thatresembles bedrock relief and soil texture distribution within the soil matrix supportthis hypothesis. Research by Michael (1982), Löffler et al (1984), and Kubiniok (1990)revealed strong relationships among regional tectonics, sedimentation, climatic con-ditions, and soil genesis. Most soils of the Korat Plateau more likely developed on anerosional relief over different rock formations. Two relief generations, which haveformed under different climatic conditions, can be distinguished. Initially, Early Ter-tiary Red Soil (Oxisols) covered most of the plateau. These soils developed underhumid tropical conditions and are currently found as isolated remnants, mainly onhills. Subsequently, increasing Miocene tectonic activity destroyed the old relief.Furthermore, the climate became increasingly drier with the glacial maximum of theLate Pleistocene. Consequently, yellowish, brownish soils (Oxisols, Ultisols, Alfisols)developed on the now gently sloping landscape. Quaternary weakly developed allu-vial soils (Entisols and Inceptisols) are found in the alluvial plains of the Mun River


and its tributaries (Kubiniok 1990). As these geomorphologic processes led to theformation of soils with distinctively different characteristics, knowledge of soil-land-scape relationships helps in a better appreciation of variation in soil properties foragronomic decision making.

This is illustrated by an example using a soil texture classification for a subareaof the Ubon Ratchathani Land Reform Area in Ubon Ratchathani Province of north-east Thailand (the rectangular area in Figure 1). A soil map (Fig. 2A) of northeastThailand at 1:50,000 scale (Changprai et al 1971) was used to derive a thematic mapof soil texture by recoding the soil classes with their associated texture classes (Fig.2B). Using this method, 93% of the mapped area would be assigned to a medium-textured class. A very different picture (Fig. 2C) emerges when soil texture was mappedusing improved interpolation techniques (see detailed explanation below). This mapclearly shows short-range variation in soil texture, with 40%, 36%, and 24% of thearea under light, medium, and heavy texture, respectively. Insufficient spatial resolu-tion, resulting in incorrect allocation of land to soil texture classes, however, rendersthe recoded soil map unsuitable to support agronomic decisions at the subregionalscale.

Can we then use existing soil classification systems as a basic source of soilinformation in the rainfed rice lands in northeast Thailand? Use would require im-proving the existing soil maps to obtain spatial estimates of those land characteristicsthat are not yet well predicted. The actual decision as to whether revision or upgrad-ing is the preferred choice depends on the quality of the existing database, the type ofproblems to be solved, and available funds (Brus et al 1992). Collection of additionalquantitative data is tedious and costly. Collecting qualitative, auxiliary information isless costly, but this information has been infrequently used in the past, perhaps be-cause of the lack of appropriate data management techniques and a research approachthat emphasizes quantitative data. The knowledge of local experts is particularly ben-eficial because of their strong bonds with the local environment. The identification ofmethods that integrate sparse quantitative and available auxiliary data is a prerequi-site for producing meaningful soil information at a regional scale in the heteroge-neous rainfed lands of northeast Thailand.

Understanding soil variability

Data needsA case study was carried out within the 61,000-ha Ubon Ratchathani Land ReformArea (URLRA), located about 45 km southeast of Ubon Ratchathani City (Fig. 1) inthe southwestern corner of the Korat Plateau, which is the largest plateau in SoutheastAsia extensively used to grow rice (Mackill et al 1996). The URLRA, which is ad-ministered under the Thailand Agricultural Land Reform Act, straddles two districts,Amphoe Det Udom and Amphoe Nachaluai, of Ubon Ratchathani Province, and isbordered by the Lam Dom Yai River. The starting point would be a set of point-basedsoil data and field observations that would normally be collected in a soil survey.Previous studies have mapped soil series as a result of semidetailed surveys (Changprai


Fig. 1. The project region in the Ubon Ratchathani Land Reform Area (URLRA) in northeastThailand.

Sampling sites:GridTransectRandom

Soil types

(Changprai et al 1971)

Alluvial Complex

Korat

Korat/Phon Phisay

Nam Phong

Phen

Phon Phisay

Roi Et

Roi Et/Phen

Ubon

Subarea

0 2 4 6 8 km


Fig. 2. Mapping soil texture classes. Map A shows the soil map of Changprai et al (1971) for anarea in the Ubon Ratchathani Land Reform Area (URLRA). Map B is the soil texture class mapderived by recoding map A. Map C was generated using advanced interpolation techniques(soft indicator kriging and classification).

et al 1971) but the accompanying point information that is important for agronomicland assessment and management was not accessible for us. Therefore, we designed asurvey scheme to describe the spatial distribution of agronomically important soilproperties at greater resolution. The development of methodologies for regional char-acterization was the main objective in this study, and within-field variability was notconsidered in the survey. It is important to note that the sampling density of thissurvey was sufficient to produce soil maps at scales of 1:50,000 to 1:100,000. Appli-cation of the proposed method in areas that have been mapped at these scales willhence require only a reanalysis of existing data and collecting some complementarydata.

Methodology for data collection and processingThe sampling layout (Fig. 1) comprised three different but complementary sets. First,a sampling grid was drawn on a map and the fields nearest to the grid nodes wereidentified using a global positioning system (Set 1). Spacing between the grid nodes

A. Soil map 1:50,0000(Changprai et al 1971)

B. Soil texture class map(reclassified soil map)

C. Improved soil texture classmap (soft indicator krigingof field texture)

Alluvial Complex soilsKorat seriesKorat/Phon Phisay associationNam Phong seriesPhen seriesPhon Phisay seriesRoi-Et seriesRoi-Et/Phen associationUbon series

Light textureMedium textureHeavy textureNo data

Light textureMedium textureHeavy texture

0 2 4 6 8 10 km


was approximately 2,000 m with a total of 91 sampling locations in the area. Posi-tional deviations from the original grid node locations were due to site inaccessibility.Second, to depict local variations, 117 sampling sites were located within 29 transects(Set 2). Each transect comprised three to five sampling locations with 50 m to 150 mspacing. Transects were positioned along the slopes of micro-catchments. Third, avalidation set (Set 3) with 70 locations was obtained using a stratified random sam-pling design. The area was split into 14 strata of equal size and five locations wererandomly selected for each stratum. For all three sampling sets, each sampling siterepresented a rice field. Soil samples were collected from 0–0.15-m depth (topsoil;soil monolith 0.2 × 0.2 × 0.15 m by spade) and from 0.15–0.4-m depth (subsoil;Dutch auger with 0.1-m diameter). Five soil samples were bulked, one from the cen-ter of the field and four samples within a 6-m radius around the center of the field.The samples were air-dried, ground to pass through a 2-mm sieve, and analyzed (Table1). Qualitative soil profile descriptions were conducted on auger borings at each loca-tion.

Where possible, interviews were conducted with the farmers of the sampledfields to record their perceptions on soil fertility of their fields (ranked on an ordinalscale: very low, low, medium, high, very high), the probability that water is sufficientfor rice production (two, four, six, eight, or ten out of ten years), and the likelihood ofiron toxicity (high, medium, low, very low, none). Interviews also revealed informa-tion about the field’s topographical position in the toposequence (low, medium, high).Farmers make use of this information and combine it with knowledge about the per-formance of different rice varieties to build strategies to minimize risks of crop fail-ure. Survey observations suggested that fields with an upper sandy horizon deeperthan 1.5 m and fields frequently rejuvenated by alluvial sediments of the Lam DomYai River should be added to the farmers’ field classification (FFC) of the three topo-graphical positions. This updated classification is referred to as the UFFC. Exhaus-tive coverage with the UFFC was achieved in an 18,000-ha subarea of the URLRA,delineated by a rectangle in Figure 1. Scanned, pan-chromatic aerial photographs(1:4,000) for this subarea were provided by the Thailand Agricultural Land ReformOffice. Within a GIS, a 250 × 250-m grid was generated for the subarea and one of theUFFC classes was assigned to each of the 2,864 grid nodes, visually aided by thescanned aerial photographs and an elevation map. The procedure was validated in1997 when locations of 100 randomly selected grid nodes were visited in the field.Classification accuracy of 96% was obtained.

Summary statisticsAbout 50% of the samples had more than 70% sand and less than 7% clay in thetopsoil. Clay contents increased slightly in the subsoil. However, soil texture wasloamy sand or loam in many parts of the study region because silt contents reachedalmost 20% in about 50% of the samples (topsoil and subsoil). About 25% of alltopsoil samples had organic carbon contents of more than 8.2 g kg–1. Subsoil organiccarbon values were very low, however, with three-quarters of the samples having lessthan 4.4 g kg–1. Bray-II P content was more than 1.1 mg kg–1 and more than 4 mg


Tabl

e 1

. Sta

tist

ics

of t

opso

il (0

–15

cm

) pr

oper

ties

in t

he a

rea

(n =

278).

Mea

ns, v

aria

nces

, and

coe

ffic

ient

s of

var

iati

on (

CVs)

wer

e es

tim

ated

by

the

met

hod

of m

omen

ts. Ap

aC

layb

Silt

bS

andb

pHc

ECd

OC

ePf

CEC

gC

ahM

ghK

hN

ah

(cm

)(%

)(%

)(%

)(d

S m

–1)

(g k

g–1)

(mg

kg–1

)(c

mol

(cm

ol(c

mol

(cm

ol(c

mol

kg–1

)kg

–1)

kg–1

)kg

–1)

kg–1

)

Tops

oil

Mea

n1

39

.42

0.5

70

.14

.80

.24

7.0

5.2

2.1

80

.31

0.1

10

.03

90

.07

Min

imum

61

.00

.12

3.3

3.9

0.1

10

.50

.60

.24

0.0

10

0.0

01

0Lo

wer

qua

rtile

12

5.3

12

.16

0.9

4.6

0.1

75

.13

1.0

40

.11

0.0

70

.01

50

.03

Med

ian

13

7.2

18

.27

3.1

4.7

0.2

16

.64

1.6

80

.19

0.0

90

.02

80

.05

Upp

er q

uart

ile1

51

0.4

27

.98

2.2

5.0

0.2

78

.26

2.4

00

.35

0.1

30

.05

00

.07

Max

imum

21

48

.45

4.1

92

.56

.41

28

.15

61

3.1

23

.37

0.5

90

.21

70

.63

CVi

18

77

51

21

84

24

79

28

91

36

65

94

12

4

Sub

soil

Mea

n1

3.1

19

.26

7.7

5.2

0.2

13

.41

.81

2.8

40

.53

0.1

30

.04

30

.19

Min

imum

0.7

02

2.7

4.2

0.1

00

0.5

00

.04

0.0

10

.02

0.0

01

0Lo

wer

qua

rtile

5.8

12

.35

6.2

4.9

0.1

32

.10

.97

0.9

60

.10

0.0

70

.01

10

.03

Med

ian

10

.51

7.9

70

.35

.10

.16

3.0

1.0

91

.92

0.2

20

.10

0.0

28

0.0

6U

pper

qua

rtile

16

.12

6.1

80

.55

.30

.20

4.4

23

.68

0.5

10

.15

0.0

57

0.1

2M

axim

um5

6.5

46

.39

6.7

8.1

0.9

71

4.3

14

15

.36

7.3

61

.48

0.4

12

4.8

0C

Vi7

84

82

49

77

68

49

41

64

88

12

12

68

a Est

imat

ed o

n au

ger b

orin

gs. b

Hyd

rom

eter

met

hod.

c At 1

:1 s

oil t

o so

lutio

n (w

ater

), m

easu

red

with

a g

lass

ele

ctro

de. d

Sat

urat

ion

extr

act.

eW

alkl

ey a

nd B

lack

wet

oxi

datio

n (B

lack

1965, p

1372-1

376).

f 0.1

N H

Cl +

0.0

3 N

NH

4F

at 1

:10 s

oil t

o so

lutio

n fo

r 4

0 s

(B

ray

and

Kur

tz 1

94

5).

g1

N a

mm

oniu

m a

ceta

te a

t pH

7 u

sing

a s

team

dis

tilla

tion

tech

niqu

e(P

age

1982, p

893

-895).

h1 N

am

mon

ium

ace

tate

at

pH 7

usi

ng a

1:5

soi

l to

solu

tion

(Pag

e 1

98

2,

p 1

59

-16

5).

i In

%.


kg–1 in 50% of the topsoil and subsoil samples, respectively. Contents of CEC andexchangeable bases were low in a large proportion of the study area. Nevertheless,1N NH4-acetate extractable K, for example, was more than 0.06 cmol kg–1 in at least25% of all samples (Table 1).

Statistical analysis of the soil samples underlined that accounting for variabilityis important in heterogeneous rainfed rice lands. Only a few soil properties had littlevariability within the study region. Many topsoil and subsoil properties related tonutrient availability (Bray-II P, CEC, exchangeable bases) had large CVs (Tables 1and 2). Even properties that are considered relatively stable, such as organic carbonor clay content (Oberthür et al 1996), had high CVs. Compared with results publishedin the seminal review of Beckett and Webster (1971), CVs found for soil attributessuch as texture, organic carbon content, or the base complex were generally muchhigher than the proposed yardstick values. Even coefficients of variation determinedby Davis et al (1995) for soils similar to the soils in northeast Thailand were onaverage lower than the CVs presented here. However, we ask the reader to interpretthe CV values with caution as the values depend on the mean of the sample. A set ofobservations with a mean close to zero has a larger CV than a set of observations ofthe same property with a higher mean. The high CVs of the bases are partly caused bythis effect.

The UFFC incorporates local changes in topography, yet CVs of many soilproperties were only slightly lower than those based on soil series (Table 2), suggest-ing that steep gradients of change exist over very short distances. But unlike the mappedseries of the 1971 soil map, the units of the UFFC have clear differences in their soilfertility properties. Grouping the individual samples using the UFFC shows charac-teristics of extremely unfavorable to favorable soil fertility. Differences between theclass medians were large for most soil properties, except for soil pH, EC, and Bray-IIP in the topsoil and pH and EC in the subsoil (Table 3). Class medians of soil fertilitycharacteristics, including organic carbon, texture, CEC, and exchangeable bases, sug-gest that soil productivity in the region declines in the order alluvial fields > fields inlow topographic position > fields in medium and high topographic position > fieldswith upper, deep sandy horizons.

Table 4 shows that the proportion of variance in the data explained by the soilmap, as measured by the complement of the relative variance RVc (Webster and Oliver1990), was low for all the soil variables determined. The RVc ranged from 0% to 8%for topsoil and subsoil, even for relatively stable soil properties such as texture, or-ganic carbon, and CEC. This illustrates the inability of the soil map to account formuch of the variance of agronomically important soil properties (Table 4). In otherwords, the mapped soil units are highly variable in their indicators of soil fertility andare of limited use for agronomic interpretation. On the other hand, the higher RVcvalues obtained with the UFFC classification (ranging from 0% to 43% for both soildepths) show that this classification accounted better for the variability of the agro-nomically important soil properties, with the exception of EC and K content, than theconventional soil mapping units (Table 4). Relatively high RVc values were achievedfor soil properties such as Ca and Mg.


Tabl

e 2

. The

coe

ffic

ient

of v

aria

tion

(%

) w

as c

alcu

late

d fo

r fo

ur m

appi

ng u

nits

of t

he s

oil m

ap o

f Cha

ngpr

ai e

t al

(1971)

and

for

five

cate

gories

of t

heup

date

d fa

rmer

s’ fie

ld c

lass

ifica

tion

(U

FFC

).

Map

ping

uni

ts o

f 1971 s

oil t

axon

omic

map

aC

ateg

orie

s of

the

UFF

Cb

KO

KO

/PP

RE

RE/

PHAB

SLT

PM

TPH

TPD

US

HS

oil

laye

rTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ub

Cla

y7

07

88

07

69

18

66

87

05

64

87

06

45

56

87

37

03

07

8S

ilt5

14

45

45

05

25

83

53

92

54

63

33

65

74

94

94

83

54

9S

and

22

24

20

22

32

37

21

23

25

28

24

28

20

20

16

16

51

0pH

99

89

51

25

16

64

810

810

98

77

EC3

17

94

57

35

53

35

98

92

81

65

08

03

27

23

66

92

81

09

OM

44

57

35

59

57

40

31

53

50

59

45

48

36

60

41

54

46

73

P1

14

81

79

93

29

77

73

83

60

84

87

64

55

70

78

96

16

12

92

CEC

75

93

96

10

18

08

98

15

16

95

87

57

45

98

57

81

18

49

87

Ca

10

81

44

12

71

64

92

12

01

00

17

79

71

15

88

10

17

31

93

18

22

30

70

17

2M

g6

41

03

40

49

69

86

44

23

53

48

54

60

47

57

71

13

53

54

1K

95

11

98

91

37

11

41

33

76

71

41

68

78

79

10

41

38

96

11

31

07

12

2N

a1

17

27

31

00

26

91

27

19

51

30

23

31

33

72

10

12

14

13

52

43

14

13

22

73

89

a KO

= K

orat

(n

= 9

8);

KO

/PP

= K

orat

/Pho

n Ph

isay

(n

= 1

22);

RE

= R

oi E

t (n

= 1

2);

RE/

PH =

Roi

Et/

Phen

(n

= 1

5).

Top

= t

opso

il, S

ub =

sub

soil.

bAB

S =

allu

vial

site

s on

the

back

slop

es o

f the

Lam

Dom

Yai

Riv

er (n

= 9

); L

TP =

low

top

ogra

phic

pos

ition

(n =

69

); M

TP =

med

ium

top

ogra

phic

pos

ition

(n =

10

7);

HTP

= h

igh

topo

grap

hic

posi

tion

(n =

65

);D

US

H =

upp

er s

andy

hor

izon

dee

per

than

1.5

m (

n = 2

3).


Tabl

e 3

. Cla

ss m

edia

ns w

ere

calc

ulat

ed fo

r fou

r map

ping

uni

ts o

f the

soi

l map

of C

hang

prai

et

al (

1971)

and

for f

ive

cate

gories

of t

he u

pdat

ed fa

rmer

s’fie

ld c

lass

ifica

tion

(U

FFC

).

Map

ping

uni

ts o

f 1971 s

oil t

axon

omic

map

aC

ateg

orie

s of

the

UFF

Cb

KO

KO

/PP

RE

RE/

PHAB

SLT

PM

TPH

TPD

US

HS

oil

laye

rTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ubTo

pS

ub

Cla

y8

.00

10

.17

6.2

09

.20

8.8

01

4.0

07

.47

12

.50

13

.60

14

.80

10

.70

15

.60

6.8

09

.85

6.6

67

.63

3.9

03

.80

Silt

18

.09

17

.93

18

.20

16

.29

18

.90

19

.20

21

.93

20

.07

21

.90

18

.30

24

.10

24

.10

17

.93

16

.50

15

.37

14

.29

11

.00

12

.10

San

d7

3.5

57

1.0

97

4.9

07

1.2

06

8.1

06

5.8

07

0.7

46

5.5

26

7.6

06

7.4

06

3.9

05

6.7

37

5.0

17

1.3

07

7.4

97

6.7

08

6.1

08

6.2

0pH

4.7

65

.04

4.7

95

.13

4.7

15

.05

4.7

05

.22

4.8

85

.06

4.8

45

.24

4.7

45

.09

4.7

25

.08

4.5

94

.86

EC0

.22

0.1

60

.20

0.1

60

.21

0.1

70

.27

0.1

70

.21

0.1

60

.25

0.1

70

.20

0.1

60

.20

0.1

60

.20

0.1

3O

M0

.72

0.3

30

.60

0.2

60

.54

0.3

00

.60

0.3

41

.17

0.5

30

.69

0.3

00

.59

0.3

00

.66

0.2

80

.57

0.3

0P

4.0

01

.53

4.0

01

.00

4.0

01

.00

4.0

01

.00

6.8

62

.04

4.0

01

.00

4.0

01

.53

4.0

01

.84

4.0

02

.65

CEC

1.8

81

.92

1.3

61

.76

1.7

61

.84

1.7

62

.96

4.1

65

.04

2.3

23

.04

1.4

42

.08

1.5

21

.28

1.0

40

.72

Ca

0.2

40

.22

0.1

60

.21

0.1

80

.23

0.1

60

.32

0.5

50

.52

0.3

40

.51

0.1

60

.20

0.1

60

.16

0.0

70

.06

Mg

0.1

10

.13

0.0

80

.09

0.0

90

.09

0.0

80

.10

0.2

20

.21

0.1

20

.13

0.0

80

.10

0.0

90

.10

0.0

70

.07

K0

.03

0.0

30

.03

0.0

20

.03

0.0

30

.03

0.0

40

.09

0.0

80

.04

0.0

40

.03

0.0

30

.03

0.0

20

.02

0N

a0

.05

0.0

50

.05

0.0

60

.04

0.0

60

.06

0.0

90

.05

0.0

60

.07

0.1

20

.04

0.0

70

.04

0.0

30

.04

0.0

5

a KO

= K

orat

(n

= 9

8);

KO

/PP

= K

orat

/Pho

n Ph

isay

(n

= 1

22);

RE

= R

oi E

t (n

= 1

2);

RE/

PH =

Roi

Et/

Phen

(n

= 1

5).

Top

= t

opso

il, S

ub =

sub

soil.

bAB

S =

allu

vial

site

s on

the

back

slop

es o

f the

Lam

Dom

Yai

Riv

er (n

= 9

); L

TP =

low

top

ogra

phic

pos

ition

(n =

69

); M

TP =

med

ium

top

ogra

phic

pos

ition

(n =

10

7);

HTP

= h

igh

topo

grap

hic

posi

tion

(n =

65

);D

US

H =

upp

er s

andy

hor

izon

dee

per

than

1.5

m (

n = 2

3).


Table 4. The complement of the relative variance (%) calculated forfour mapping unitsa of the soil map of Changprai et al (1971), forthree categoriesb of the farmers’ field classification (FCC) and forfive categoriesc of the updated farmers’ field classification (UFFC).

Soil mapa FCCb UFFCc

Topsoil Subsoil Topsoil Subsoil Topsoil Subsoil

Clay 3 1 10 17 21 24Silt 0 0 16 15 25 17Sand 0 1 18 24 30 29pH 0 2 0 2 0 4EC 4 0 8 1 8 0OM 4 0 0 0 25 15P 0 1 0 3 2 3CEC 0 1 17 15 29 20Ca 0 1 23 11 34 13Mg 8 5 0 4 43 20K 3 0 32 3 0 11Na 2 1 4 2 7 2

aKorat, Korat/Phon Phisay, Roi Et, Roi Et/Phen. bLow, medium, high position.cLow, medium, high position, deep sands, alluvial backslope sites.

Mapping soil variability

Managing soil variability requires knowledge about the spatial distribution of soilproperties. Geostatistical interpolation approaches permit the user to first elucidatethe spatial structure of soil properties using variography (i.e., examining plots of thesemivariance, which is a measure of the dissimilarity of pairs of data points, againstdistance between them), and second to use this information in interpolation algo-rithms to generate thematic soil maps (Oliver and Webster 1991). The indicator ap-proach (Goovaerts 1997) was chosen because it accounts for class-specific patternsof spatial continuity through the different indicator variogram models at each thresh-old, allows for the incorporation of additional (secondary soft) information, and doesnot depend on normality of the data. Qualitative and quantitative data can be pro-cessed using the indicator approach. In the case of quantitative data, one is able todetermine the conditional variance σ2 of the data around their mean. Unlike the vari-ance obtained for nonindicator approaches, the conditional variance depends on theactual data and not on the covariance model. Nonindicator-based kriging varianceswould be the same for a given covariance model and data configuration even whendifferent data were used to derive estimates of a variable Z.

Soft indicator kriging (SIK) and indicator simulations (IS) were used for theinterpolation. The flow of operations for this study is outlined in Figure 3. Krigingalgorithms are best locally, whereas simulation models the joint uncertainty aboutattribute values at several locations. Connected strings of large and small values are


Hard point data

Clay, silt, sand

Hard point data

Soil texture classes

Soft point data

UFFC classes

Area information

Aerial photography

Geographic information systems

Soft grid data

UFFC classes

Sequential ISMaps of clay, silt, sand

SIKMaps of texture classes

Geostatistical indicator interpolation

better reproduced by simulation than kriging, and this is important for an assessmentof soil property distribution in very heterogeneous lands.

Suppose that {z*(u), uεA)} is a set of kriging estimates of the soil property Z forlocations u in a region A. The local error variance Var{Z*(u) – Z(u)} is minimum andthe best estimate in the “least-square sense” if each estimate z*(u’) is consideredindependently of its neighboring estimates. These best local estimates may not bebest as a map and, instead of generating such a kriged map, simulation produces oneor a set of possible realizations of the spatial distribution of the z* values. For ex-ample, a map of z* values {z* (l)(u), uεA}, where l denotes the lth realization, can begenerated that reproduces the statistics of an area that are most important for theproblem at hand. Simulated maps honor data values at their locations and reproduceclosely the declustered sample histogram and the set of covariance models for vari-ous indicator thresholds (Goovaerts 1997).

In this study, indicator variograms were modeled for the nine deciles of quanti-tative soil properties and for the classes of qualitative properties. Estimates of mean(Figs. 4A–6A), median (Figs. 4B–6B), conditional variance (Figs. 4E–6E), and con-ditional CV (Figs. 4F–6F) and the 0.2- and 0.8-quantiles (Figs. 4C,D–6C,D) of theconditional cumulative distribution function were mapped. Quantile maps are oftenmore useful than maps showing mean values; for instance, overestimation of soilnutrients in regions with low soil fertility is more harmful (yield loss or crop failure)than underestimation (undue application of treatments). Conversely, underestimationof nutrient contents has negative impacts in very fertile regions. In this situation,excess fertilizer nutrients are discharged in the groundwater and have detrimentaleffects on ecosystems. Using quantile maps, one can over- or underestimate the truevalues and minimize risks of environmental pollution or crop failure (see Figs. 4–6).

Fig. 3. The outline of operations that are needed to generate mapsfrom point data using sequential indicator simulation (IS) and softindicator kriging (SIK).


Fig. 4. Applying estimators to probabilities of occurrence (obtained using sequential indicatorsimulation) for created maps of clay (topsoil). Map A shows the mean value and was generatedusing the e-type estimator. Quantile estimators were applied to produce maps B, C, and D thatshow the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distributionfunction (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B), 80%(C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and F depictthe spread about the mean at each location as the conditional variance (map E; Con. Var.) andin relative terms as the conditional coefficient of variation (map F; Con. CV).

A. Clay (%)

C. 0.2 ccdf C. 0.8 ccdf

E. Con. Var. F. Con. CV

B. 0.5 ccdf


Fig. 5. Applying estimators to probabilities of occurrence (obtained using sequential indicatorsimulation) for created maps of silt (topsoil). Map A shows the mean value and was generatedusing the e-type estimator. Quantile estimators were applied to produce maps B, C, and D thatshow the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distributionfunction (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B), 80%(C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and F depictthe spread about the mean at each location as the conditional variance (map E; Con. Var.) andin relative terms as the conditional coefficient of variation (map F; Con. CV).

A. Silt (%) B. 0.5 ccdf

C. 0.2 ccdf D. 0.8 ccdf

E. Con. Var. F. Con. CV


Fig. 6. Applying estimators to probabilities of occurrence (obtained using sequential indicatorsimulation) for created maps of sand (topsoil). Map A shows the mean value and was gener-ated using the e-type estimator. Quantile estimators were applied to produce maps B, C, and Dthat show the values at the 0.5-, 0.2-, and 0.8-quantile of the conditional cumulative distribu-tion function (ccdf), respectively. Estimates can be interpreted as the value that has 50% (B),80% (C), or 20% (D) probability of being exceeded by the true value in the field. Maps E and Fdepict the spread about the mean at each location as the conditional variance (map E; Con.Var.) and in relative terms as the conditional coefficient of variation (map F; Con. CV).

A. Sand (%) B. 0.5 ccdf

B. 0.2 ccdf

F. Con. CV

D. 0.8 ccdf

E. Con. Var.


In summary, the chosen geostatistical indicator algorithms have various com-parative advantages over other mapping approaches, including the ability to handlequantitative and/or qualitative variables, to account for uncertainty about the esti-mated data and to offer various types of estimates (e.g., mean, median, quantiles,probability of exceeding a given threshold).

Maps were generated for all analyzed soil properties and can be obtained fromthe IRRI-GIS laboratory. In this chapter, we discuss only the geostatistical analysis oftopsoil texture classes and the mapping of topsoil particle size distribution. Soil par-ticle size distribution was mapped for the URLRA using sequential indicator simula-tion with hard data from laboratory analysis of all soil survey samples, without softinformation such as the UFFC classification. The resulting maps (Figs. 4–6) showthat average clay, silt, and sand contents have a similar spatial distribution. Clay islow in most parts of the region and moderately high only along the Lam Dom YaiRiver in the southwestern part of the study area. Silt follows similar patterns but,unlike clay, has another inclusion with higher values in the northeastern part of thestudy area. High sand contents form a wide belt that stretches diagonally from thenorthwestern to the southeastern part of the study area.

The map outputs (Figs. 4–6) differ from those generated using soft informa-tion, such as Figure 2C. The soil texture class map shown in Figure 2C, based on fieldtexturing, was generated for the rectangular subarea of the URLRA using a modifiedindicator kriging approach (soft indicator kriging or SIK) that incorporates descrip-tive soft information into the mapping procedure (Oberthür et al 1999). A probabilityvector of occurrence of each soil texture class at unsampled locations was estimatedfrom hard information (field estimations of soil texture classes) and soft informationabout field location (formalized knowledge of farmers and experts in the UFFC).

Our data sets would not have permitted meaningful kriging of soil texture classesif we hadn’t reduced the number of possible classes by merging some of those previ-ously recognized. Original topsoil texture classes as estimated in the field were ag-gregated into three major classes to obtain sufficient samples for geostatistical mod-eling in each class. Relative to the average soil texture in the study area, these classesrepresent soils with light (class one), medium (class two), and heavy (class three)texture. Consequently, class one represents sandy and loamy sand soils and class twosandy loam. All samples classified as loam and heavier are assigned to class three.

The SIK map (Fig. 2C) reveals a dendritic pattern (similar to tree branches) andmuch spatial detail of soil texture class distribution. Texture classes change graduallyfrom heavy to medium to light and follow the main topographic and drainage pat-terns. In the undulating rainfed lands of northeast Thailand, knowledge of farmersand soil experts implicitly reflected the spatial variation of soil texture classes andthis soft information was incorporated in the geostatistical approach to improve map-ping accuracy. The dendritic pattern reflects partly the local topography and drainagesystem. However, considering only topography and drainage pattern as soft informa-tion would not be sufficient because correspondence between these features and soiltexture class distribution is only partial: the size of micro-watersheds in the area issmaller than the extent of the contiguous areas of soil texture classes.


Fig. 7. The indicator variograms for three finger-estimated soil texture classes (light, medium,heavy).

The SIK map confirms the short-range variation in soil texture that is indicatedby variograms of the texture classes (Fig. 7). The variograms display mainly short-range structures, indicating a change of texture classes over very short distances.Variography results listed in Table 5 reveal that low clay values change rapidly in thetopsoil over short distances. No spatial dependence was detected beyond 1 km forvalues corresponding to indicator thresholds < 0.5 of the conditional cumulative dis-tribution function (corresponding to 7.2% clay content). Large contents of topsoilclay are spatially dependent over longer distances of up to 2.8 km. Low silt values are

Northeast Thailandfield texture—Class 1

0.25

0.20

0.15

0.00

0 1 2 3 4 5 6 7

Semivariance

8

Class 2

0.25

0.20

0.15

0.10

0.000 1 2 3 4 5 6 7 8

Distance (km)

Class 3

0.070.060.050.040.030.00

0 1 2 3 4 5 6 7 8

Table 5. Effective rangea (given in m) and the relative nugget effectb of the exponential modelsand pure nugget variation that were fitted to the indicator variogramsc: depth of clay, silt, andsand of topsoil and subsoil.

T 0.1 T 0.2 T 0.3 T 0.4 T 0.5 T 0.6 T 0.7 T 0.8 T 0.9

ClayEffective range 613 613 759 846 847 1,372 2,015 2,803 1,518Relative nugget 0.25 0.49 0.39 0.55 0.55 0.34 0.40 0.44 0.38

effect

SiltEffective range 2,131 2,453 2,803 993 788 934 1,693 1,168 –Relative nugget 0.20 0.27 0.37 0.31 0.28 0.31 0.31 0.25 1

effect

SandEffective range – 993 701 1,226 701 1,401 4,088 3,737 4,380Relative nugget 1 0.40 0.26 0.26 0.27 0.49 0.47 0.24 0.20

effect

aFor practical purposes, the effective range, i.e., three times the distance parameter of the exponential model,is given to indicate the limit of spatial dependence. bThe relative nugget effect was calculated as the ratio ofnugget variance (discontinuity at the beginning of the variogram caused by sampling error and short-scalevariability not resolved by the sampling scheme) to sill (value where the variogram reaches a plateau). cIndicatorvariograms were calculated with values for thresholds (T) corresponding to the nine deciles (0.1–0.9) of theconditional cumulative distribution function.


related over slightly longer distances than high silt contents. High sand values arespatially dependent over long distances of up to 4.4 km, whereas low sand contentschange rapidly over short distances (Table 5).

Our results are thus not consistent with the general belief that soils of the KoratPlateau are universally coarse-textured and infertile. The SIK map assigns much land(25%) to soil texture class three (loam and heavier). A loamy soil is expected to besuitable for rice production. The reliance on soil maps may have contributed to thelong-held views about soil texture in northeast Thailand. In reality, soil texture in thisregion varies over short distances and the soil maps do not depict these changes.Colluvio-alluvial processes over short distances and in situ soil development are asimportant factors for the distribution of soils in northeast Thailand as is the graduallychanging Quaternary alluvial sedimentation.

Conclusions

1. Soil maps at scales of 1:100,000 to 1:125,000 highlight soil trends in theregion but don’t depict much of the agronomically important spatial vari-ability and cannot be used for land resource assessment. Agronomic inter-pretations should be based on maps generated by detailed soil surveys or oncombining field data with existing information to produce refined thematicmaps with relevant spatial resolution.

2. The UFFC that is based on local knowledge captures much of the variationin soil properties and does provide a basis for land resource assessment. Itsapplicability, however, excludes some soil properties (see RVc values) andlittle variation of pH, EC, Bray-II P, K, and Na is accounted for. Variation ofmore stable soil properties characterizing the inherent soil potential (texture,organic carbon, CEC) and Ca and Mg of the base complex can be assessedsufficiently.

3. Although the average values of soil fertility indicators are low, their rangeand spatial distribution, in conjunction with soil development processes, con-firm information by farmers that soil fertility levels are sufficient for rainfedrice production in a large proportion of the land.

4. The soft information used here is readily available in many regions or doesnot cost much to obtain. Therefore, instead of concentrating on additionalsampling with greater density, SIK provides an interesting cheap alternativefor updating or upgrading soil maps and describing spatial soil variability.

5. Integration of hard and soft information revealed visually that colluvio-allu-vial processes act over short distances and in situ soil development is likely.Large-scale variation, as suggested by the variograms and maps, is not con-sistent with the widely held belief that Quaternary alluvial sedimentationwas the sole process responsible for the distribution of soils in the region.

6. Geostatistical indicator approaches offer a suite of tools for characterizingand mapping soil variability in the rainfed rice lands, including variography


and the possibility to generate quantile maps or maps that exceed specifiedthresholds.

7. The results presented indicate the need for software that implements the SIKand other relevant algorithms in a standardized and user-friendly mode.

ReferencesBeckett PHT, Webster R. 1971. Soil variability: a review. Soils Fert. 34:1-15.Black CA. 1965. Methods of soil analysis. Madison, Wis. (USA): American Society of

Agronomy.Bray RH, Kurtz LT. 1945. Determination of total, organic, and available forms of phosphorus

in soils. Soil Sci. 59:39-45.Brus DJ, De Gruijter JJ, Breeuwsma A. 1992. Strategies for updating soil survey information:

a case study to estimate phosphate sorption characteristics. J. Soil Sci. 43:567-581.Changprai C, Chotimon A, Thunduan V, Thipsuwan C, Lepananontha J, Kittiyarak S. 1971.

Detailed reconnaissance soil map of Ubon Ratchathani Province. Bangkok (Thailand):Department of Land Development.

Craig IA. 1988. Agronomic, economic and socially sustainable strategies for soil fertility man-agement in northeast Thailand. In: Chareonwatana T, Rambo AT, editors. Sustainablerural development in Asia. Fourth SUAN Regional Symposium on Agroecosystem Re-search. Khon Kaen (Thailand): Farming Systems Research Project and Southeast AsianUniversities Agroecosystem Network. p 153-165.

Davis JG, Hossner LR, Wilding LP, Manu A. 1995. Variability of soil chemical properties intwo sandy, dunal soils of Niger. Soil Sci. 159:321-330.

Dobermann A, White P. 1999. Strategies for nutrient management in irrigated and rainfed low-land rice systems. Nutr. Cycl. Agroecosyst. 53:1-18.

Goovaerts P. 1997. Geostatistics for natural resources evaluation. New York (USA): OxfordUniversity Press.

Grandstaff TB. 1988. Environment and economic diversity in Northeast Thailand. In:Chareonwatana T, Rambo AT, editors. Sustainable rural development in Asia. FourthSUAN Regional Symposium on Agroecosystem Research. Khon Kaen (Thailand): Farm-ing Systems Research Project and Southeast Asian Universities Agroecosystem Net-work. p 11-22.

Kubiniok J. 1990. Relief- und Bodengeneration auf dem Khorat-Plateau (NE-Thailand).Zeitschrift für Geomorphol. 34:149-164.

Löffler E. 1984. Quarternary geomorphological development of the Lower Mun River Basin,North East Thailand. Catena 11:321-330.

Mackill DJ, Coffman WR, Garrity DP. 1996. Rainfed lowland rice improvement. Manila (Phil-ippines): International Rice Research Institute. 242 p.

Michael P. 1982. The landforms of Thailand: ideas about their genesis and influence on soilproperty distribution. In: Nutalaya P, Karasudhi P, Tanasuthipitak T, Kheoruenromme I,Sudhiprakarn A, editors. First International Symposium on Soil, Geology and Land-forms: Impact on Land Use Planning in Developing Countries. Bangkok (Thailand):Association of Geoscientists for International Development. p C10.1-C10.14.

Mitsuchi M, Wichaidit P, Jeungnijnirund S. 1989. Soils of the Northeast Plateau, Thailand.Tsukuba (Japan): Tropical Agriculture Research Center.


Miura K. 1990. Genetic features of the major soils in northeast Thailand. Khon Kaen (Thai-land): Agricultural Development Research Project in Northeast Thailand.

Moormann FR, Montrakun S, Panichapong S. 1964. Soils of northeastern Thailand: a key totheir identification and survey. Bangkok (Thailand): Land Development Department.

Oberthür T, Dobermann A, Goovaerts P. 1999. Mapping soil texture classes using field textur-ing, particle size distribution, and local knowledge by both conventional and geostatisticalmethods. Eur. J. Soil Sci. 50:459-479.

Oberthür T, Dobermann A, Neue HU. 1996. How good is a reconnaissance soil map for agro-nomic purposes? Soil Use Manage. 12:33-43.

Oliver MA, Webster R. 1991. How geostatistics can help you. Soil Use Manage. 7:206-218.Page AL. 1982. Methods of soil analysis. Madison, Wis. (USA): American Society of Agronomy.Paiboon P, Liengsakul M, Engkagul V. 1985. Grain size analysis of some sand rises and stream

sediments in the northeast of Thailand in order to indicate depositional environment. In:Proceedings of the Conference on Geology and Mineral Resource Development in North-east Thailand. Khon Kaen University, Khon Kaen, Thailand. p 235-253.

Patcharapreecha P. 1988. Physico-chemical properties of the Northeast paddy soils in relationto fertility. In: Panichapong S, editor. Proceedings of the First International Symposiumon Paddy Soil Fertility. Paddy soil fertility working group, Chiang Mai, Thailand.p 405-414.

Pendleton RL, Montrakun S. 1960. The soils of Thailand. In: Proceedings of the 9th PacificScience Congress. Bangkok (Thailand): Department of Rice, Ministry of Agriculture.

Ragland J, Boonpuckdee L. 1988. Soil fertility management in northeast Thailand. Khon Kaen(Thailand): Northeast Rainfed Agricultural Development Project (NERAD).

Suddhiprakarn A, Kheoruenromne I. 1987. A study on some alfisols and ultisols in ustic soilmoisture regime, northeast Thailand. Kasetsart J. (Nat. Sci.) 21:214-229.

Tamura T. 1986. Geomorphological development in northeast Thailand with reference to prob-lem soil formation. Agricultural Development Research Project in Northeast Thailand.Khon Kaen (Thailand): JICA.

Webster R, Oliver MA. 1990. Statistical methods in soil and land resource survey. Oxford(UK): Oxford University Press.

Zeigler RS, Puckridge DW. 1995. Improving sustainable productivity in rice-based rainfedlowland systems of South and Southeast Asia. GeoJournal 35:307-324.

NotesAuthors’ addresses: Thomas Oberthür, Centro Internacional de Agricultura Tropical (CIAT),

Cali, Colombia; Suan Pheng Kam, International Rice Research Institute, Los Baños,Philippines.


Identifying soil suitability for subsoil compaction . . . 97

For coarse-textured soils of high subsoil permeability, research has demon-strated the benefits of subsoil compaction for improved water- and nutrient-use efficiency in rainfed lowland rice (Oryza sativa L.). To better define soilconditions suited to this subsoil compaction, on-farm experiments were car-ried out on soils varying in subsoil clay content. Rice (cv. KDML105 in 1993and cv. RD6 in 1994) was grown in main plots comparing shallow dry tillagewithout compaction, shallow dry tillage with subsoil compaction, and deepdry tillage with subsoil compaction. Soil was compacted in seven farmers’fields in the south of Ubon Ratchathani Province, with 10 passes of a vibrat-ing road roller on 19 and 23 May 1993. The effects of subsoil compaction onchanges in soil physical and hydrological properties differed according tosubsoil clay content, which ranged from 1.4% to 12.0%. Subsoil compactiondecreased soil hydraulic conductivity sufficiently for fields with subsoil claycontent greater than 2%. When subsoil clay content was higher than 10%,the justification for using relatively costly subsoil compaction was question-able, as the hydraulic conductivity was already low and the gains in water-storing capacity seemed limited. Based on these results, proportions of soilswith subsoils of <2%, 2–5%, 5–10%, or >10% clay were mapped using geo-graphic information systems for an area of about 40,000 ha within the UbonRatchathani Land Reform Area. About 40% of the mapped area had subsoilswith clay % between 2 and 10, suggesting that substantial areas could besuitable for subsoil compaction. More investigations are needed to assessits economic and social acceptability, to better understand variability andprobability of response, and to further refine soil suitability in relation to claytype, clay content, and groundwater changes at the toposequence level.

Identifying soil suitability for subsoilcompaction to improve water-and nutrient-use efficiencyin rainfed lowland riceD. Harnpichitvitaya, G. Trébuil, T. Oberthür, G. Pantuwan, I. Craig,T.P. Tuong, L.J. Wade, and D. Suriya-Arunroj

98 Harnpichitvitaya et al

Rainfed lowland rice, growing with an average yield of only 1.5 t ha–1 on some 4.5million ha of coarse-textured and low-fertility soils in northeastern Thailand, maysuffer from water stress as soon as rains are interrupted for about 1 wk (Sharma 1992).Percolation rate can be as high as a few centimeters d–1. Puddling is not fully effectivedue to high sand content. Efficient management of rainwater and nutrients is a keyfactor in improving the productivity and stability of rice. On-station experiments,carried out at the Ubon Ratchathani Rice Research Center on a soil in the Ubon serieswith a clay content of 12% in the subsoil (30–60 cm), showed that water- and nutri-ent-use efficiency may be improved by reducing percolation through the creation of asubsurface barrier (Garrity et al 1992). Subsequent field-oriented research to identifypractical and cost-effective techniques for decreasing subsoil permeability showedthat multiple passes of a 12-t conventional road roller with vibration action (DynapacCA25) that compacts the subsoil to 75-cm depth was most effective (Sharma et al1995a,b, Trébuil et al 1998). As subsoil compaction was observed to concentrate riceroots in the plow layer above the compacted zone, a deeper (15–20 cm) tillage depththan the one currently achieved in farmers’ fields (7–10 cm) was found to be desir-able. All previous studies were carried out at the experiment station of the Ubon RiceResearch Center. The effects of subsoil compaction across a range of subsoil claycontents have not been investigated, and the combined effects of soil compaction andgreen manuring have not been investigated either.

The objectives of the on-farm study were1. To quantify the effects of subsoil compaction and tillage depth on changes in

sandy soil physical and hydrological properties across a range of subsoilclay contents.

2. To quantify the interactions between subsoil compaction and green manurefertilization and their effects on the productivity of rainfed lowland rice.

3. To define soil suitability for subsoil compaction by using geographic infor-mation systems (GIS) techniques applied to the Ubon Ratchathani Land Re-form Area.


Experimental siteTwo-factor on-farm field experiments were conducted in the 1993 and 1994 wet sea-sons in Ban Klang and Ban Mak Mai villages of Klang subdistrict of Det UdomDistrict, Ubon Ratchathani Province, lower northeast Thailand. The experimental sandysoils (Table 1) belonged to the Nam Phong series and were acidic, low in organicmatter content and available N, P, and K, and with a high concentration of ferrousiron in some fields.

Experimental treatmentsWe selected seven farmers’ fields for this study. The elementary experimental plotsize varied from 100 to 200 m2. Each plot was surrounded by a single levee, approxi-mately 20 cm high and 40 cm wide.


In the 1993 wet season (WS), the following three subsoil compaction and till-age treatments were imposed in each of the selected fields (but without replications ineach field):

● C0 = uncompacted dry tillage; soil was disc-plowed dry to 7–10-cm depth.● C1 = subsoil compacted by 10 passes of a 12-t Dynapac vibrating road roller

on 19 or 23 May 1993 (with soil moisture brought close to field capacity byusing a watering truck), followed by dry tillage to 7–10-cm depth.

● C2 = subsoil compacted as above but with dry tillage to 15–20-cm depth.In the 1994 WS, we selected the C0 and C1 plots of six farmers’ fields, all

having been included in the previous year’s on-farm experiment. C2 plots were notincluded because the tillage treatment did not show any effects on the monitoredparameters in 1993. Each plot was further divided into two to accommodate a split-plot design, with the main and subplot treatments as follows:

● Main plots:C0 = uncompacted with shallow (7–10 cm) dry tillage.C1 = compacted with shallow dry tillage (plots were not compacted again in1994).

● Subplots:F0 = no green manure before rice (control treatment).F1 = with green manure before rice.

Cultural practicesThe experimental cultivars were the similar and widely grown photoperiod-sensitive,medium-tall KDML105 and RD6 during the 1993 and 1994 wet seasons, respec-

Table 1. Properties of the experimental soils belonging to the NamPhong series in Det Udom District, Ubon Ratchathani Province, 1993wet season.

Soil parameter Range Mean

Particle size distribution: 0–30 cm (%)Sand 88.2–94.6 90.9Silt 3.6– 9.4 6.5Clay 1.7– 3.6 2.6

Particle size distribution: 30–60/70 cm (%)Sand 88.0–95.4 91.2Silt 3.1–10.3 6.7Clay 1.4– 4.5 2.1

Chemical parameters: 0–30 cmpH (1:1) 3.9– 5.2 4.2Organic matter content (%) 0.39–1.79 0.85Total N (%) 0.02–0.08 0.04Extractable P (Bray II, ppm) 6.1–19.0 9.8Extractable K (ammonium

acetate, pH 7, ppm) 5.0–12.6 8.6CEC (meq 100 g–1) 0.32–1.28 0.83


tively. In 1993, following dry tillage and harrowing, rice was direct-seeded at a rate of62.5 kg ha–1 on 24-25 May 1993. Rice seeds were covered by a second harrowingand, as in the surrounding farmers’ fields, no weeding was implemented because onlylimited weed infestations occurred. Later, two fertilizer applications were carried outat sowing (25-11-21 kg NPK ha–1) and panicle initiation (19 kg N ha–1).

In 1994, because green manure before rice (a combination of Sesbania rostrata,Aeschynomene afraspera, cowpea, and sword bean seeds was sown in mid-May 1994)was grown in subplots, rice was transplanted at 20 × 20-cm intervals on 29 July 1994for three fields and 8 August 1994 for another three fields. A basal application of 11kg P ha–1 was made in F1 subplots before the sowing of green manure seeds. A similarP basal application was applied to subplots without green manure before transplant-ing of rice. All subplots also received 25 kg N and 21 kg K ha–1 at 32 days aftertransplanting (DAT) and 25 kg N ha–1 again at 48 DAT.

Field-level soil and water observationsA piezometer was installed to 150-cm depth in each of the experimental fields forweekly monitoring of fluctuations in the groundwater table. We did not monitor thegroundwater table depth in each individual plot because the groundwater table depthwas governed by the regional hydrology rather than the soil management at the plotlevel. A shorter piezometer (40-cm PVC pipes with perforations in the bottom 10 cm)was installed to 30-cm depth to monitor the perched water table in each plot duringthe 1994 wet season only. Floodwater depth was regularly recorded using slopinggauges permanently installed in each subplot, for seven fields in the 1993 WS and sixfields in the 1994 WS. Volumetric soil moisture content in the 0–15-cm layer wasmeasured twice a week from the last rain until rice harvest. The Det Udom Districtmeteorological station and two rain gauges installed at the experimental sites in eachof the two villages of Klang subdistrict provided daily rainfall data.

Soil penetration resistance was measured just before flowering on 12–13 Octo-ber 1993 using a recording-type cone penetrometer (Eijkelkamp, cone base area = 1cm2) for each 5-cm-thick soil layer from 0 to 75-cm depth, for seven fields in the1993 WS only. Saturated hydraulic conductivity was also determined close to matu-rity in mid-November 1993, using the constant head method for seven fields and theon-station experiment at the Ubon Rice Research Center.

Fresh and dry weights of the green manure biomass incorporated at land prepa-ration for transplanted rice were measured 1 wk before rice transplanting in 1-m2

sampling squares replicated three times in each subplot, for five fields in the 1994WS only. In both years, grain yield of rice from harvest areas of 6 m2 was recordedand adjusted to 14% grain moisture content. Yield data were obtained from sevenfields in 1993 and six fields in 1994.

GIS-based mapping of subsoil textureSoil texture data were collected in a 2,000 × 2,000-m grid and at sites in transectsalong the slopes of micro-watersheds to reveal soil heterogeneity over different spa-tial scales in an area of 40,000 ha within the Ubon Ratchatani Land Reform Area. The


locations for sampling were drawn a priori on a map, and the coordinates of the 278sampling locations were identified in the field using a global positioning system andthe nearest rice field sampled. Soil samples were collected from 0 to 15-cm (soilmonolith 0.2 m × 0.2 m × 0.15 m by spade) and 15 to 40-cm (dutch auger) depths.Five samples were bulked, one from the center of the field and four samples within a6-m radius around the center of the field. The samples were air-dried and ground topass through a 2-mm sieve. Soil particle size distribution was determined using thehydrometer method.

From the collected point-data, maps of subsoil textures were generated for eachof the 250 × 250-m cells of the 40,000-ha area using indicator geostatistics with andwithout prior means (Goovaerts 1997), as described in Oberthür et al (1999). TheGIS-based map would allow us to classify and map out cells according to their sub-soil clay contents.


Rainfall and water table depthFigure 1 displays the extent of rainfall variability during the wet season across yearsfor the Det Udom area. Seasonal rainfall distributions and groundwater table fluctua-tions during the 1993 and 1994 growing seasons contrasted markedly (Fig. 2). Totalrainfall received by the rainfed lowland rice crops was 1,035 mm for direct-seededrice in 1993 and 751 mm for the transplanted rice crop during the 1994 wet season. Inthe 1994 WS, 650 mm of rainfall were received from May to late July, when the greenmanure plants were growing before the rice crop. In 1993, significant rainshowers

Fig. 1. Cumulative rainfall in 1993 and 1994 and at probability ofexceedence of 20% (AF20), 50% (AF50), and 80% (AF80) at DetUdom, Ubon Ratchathani, lower northeast Thailand. The frequencyanalysis was based on 1963-97 rainfall record.

1,600

1,400

1,200

1,000

800

600

400

200

0

Rainfall (mm)

Jun Jul Aug Sep Oct NovMonth

AF80AF50AF2019931994


Fig. 2. Rainfall distribution, groundwater table fluctuations, and rainfedlowland rice cropping calendar during the 1993 (A) and 1994 (B) wetseasons in Det Udom District, Ubon Ratchathani Province, lower north-east Thailand.

150

100

50

0

F1 F2

TP H HAR

Days after transplanting

Rainfall (mm) Water table depth (cm)

0

–50

–100

–150–50 0 50 100 150 200

TP = transplantingF1–2 = 1st, 2nd fertilizer

applicationH = headingHAR = harvest

B

150

100

50

0 0

–50

–100

–150–20 0 20 40 60 80 100 120 140 160 180 200

Days after sowing

F1 F2 F3 H HAR

A

CB

Rainfall (mm) Water table depth (cm)

C = soil compactionB = broadcastingF1–3 = 1st, 2nd, 3rd fertilizer

applicationH = headingHAR = harvest


ceased at 116 days after sowing (DAS), apart from a small shower about 1 wk beforeheading (Fig. 2A). In 1994, rain also stopped about 1wk before heading at 70 DAT(for plots transplanted on 8 August), or 80 DAT (for plots transplanted on 29 July).During the 1994 WS, no dry spells longer than 5 d occurred between transplantingand the last rainshower before heading (Fig. 2B).

In 1993, the water table depth stayed at less than 50 cm below the soil surface inmost of the fields from 85 to 130 DAS (some 10 d after panicle initiation). Such highgroundwater table levels mitigated the effects on the rice crop of the dry spells occur-ring during that period. During the 1994 WS, the mean groundwater table depth roseto less than 20 cm from the soil surface from 15 d before transplanting and remainedat this level until 55 DAT (Fig. 2). The groundwater table remained within the top 40-cm depth up to the heading stage, but fell below 50 cm after flowering. This meansthat, in the 1994 WS, rice encountered water stress during grain filling.

Effects of subsoil compaction on soil penetration resistanceand saturated conductivityIn the seven fields for which data were collected, subsoil compaction significantlyincreased soil penetration resistance to at least 75-cm depth, with the 15–65-cm layerbeing the most compacted (Fig. 3). Subsoil compaction effectively decreased satu-rated hydraulic conductivity (by a factor of 2–3), but most of the postcompactionvalues remained higher than those commonly observed (10–20 cm d–1) in paddy fields(Fig. 4). The degree of subsoil compaction obtained on these very coarse-texturedsoils, without removing the topsoil, varied extensively, particularly according to thepercentage of clay in their subsoils. The observed effects on soil hydraulic conductiv-

Fig. 3. Soil penetration resistance profiles (mean ± SE values ofseven farmers’ fields) in different subsoil compaction and tillagetreatments, 1993 wet season, in Det Udom District, UbonRatchathani Province, lower northeast Thailand.

0

20

40

60

80

100

120

Soil depth (cm) Soil penetration resistance (kPa)

0 100 200 300 400 500

Uncompacted + shallow tillageCompacted + shallow tillageCompacted + deep tillage


ity suggested that subsoil compaction was not effective in fields with less than 2%clay in the subsoil, where the postcompaction conductivity remained exceedinglyhigh (more than 200 cm d–1). Soil hydraulic conductivity decreased significantly infields having a subsoil clay content of more than 2% but less than 5%, but not up to anoptimum of less than 10–20 cm d–1. On the other hand, where subsoil clay contentwas higher than 10%, the uncompacted hydraulic conductivity was already low. Wheresubsoil clay exceeded 10%, the technical, and probably the economic, justificationfor using subsoil compaction could be questionable, as the benefits of reducing per-colation rate were limited. Farmers’ fields with subsoil clay contents between 2% and10% seemed to be suitable for subsoil compaction. Since data points were few, thisrelationship requires further study.

Effects of subsoil compaction on field hydrological conditionsPonded water depth and duration. In the 1993 WS, no ponded water was observed inany plot before 82 DAS. Subsoil compaction did not increase the duration of flood-water, and had only a limited effect on floodwater depth during 82–120 DAS in shal-low tillage plots (Fig. 5A). In all but one field, the groundwater table was within 20–30-cm depth from 85 to 130 DAS. Shallow groundwater was a major factor maskingthe effect of subsoil compaction on floodwater duration in the 1993 WS. In the otherfield where the groundwater table was mostly below 50–100-cm depth during 80–130 DAS, the differences in floodwater depth clearly displayed the effect of subsoilcompaction on water retention above the soil surface (data not shown). Since the

Fig. 4. Relationship between percentage of subsoil clay content andsubsoil compaction effects on saturated hydraulic conductivity ofseven farmers’ fields and one field within the Ubon Rice ResearchCenter (with subsoil clay content of 12%), Ubon Ratchathani Prov-ince, lower northeast Thailand.

600

500400

300

200

100

01.4 2.1 2.3 2.6 4.4 4.6 10.9 12.0

Subsoil clay content (%)

Saturated hydraulic conductivity (cm d–1)

513

Uncompacted

Compacted

10


water table depth varied with toposequence, the findings highlight the importance ofthe toposequence for the success of subsoil compaction in prolonging the floodwaterduration.

In the 1994 WS, because of the generally high groundwater table in most fields,the total number of weeks with water ponded in the rice plots was the same for the C0and C1 treatments. The effect of subsoil compaction on floodwater duration was clearlydisplayed only in one field with a lower groundwater table for most of the wet season,where the percolation rate was higher (data not shown). In this field, the duration ofsoil submergence in the compacted C1 plots (13.0 wk) was twice as long as observed

Fig. 5. Floodwater depth (positive values) and perched water table(negative values) as affected by subsoil compaction and tillage treat-ments in (A) 1993 wet season (WS) and (B) 1994 WS, Det UdomDistrict, Ubon Ratchathani Province, lower northeast Thailand. Datawere from seven farmers’ fields in 1993 and six farmers’ fields in1994.

25

20

15

10

5

0

Floodwater depth (cm)

80 90 100 110 120

Days after sowing

A Uncompacted + shallow tillageCompacted + shallow tillageCompacted + deep tillage

Floodwater depth (cm)

50

40

30

20

10

0

–10

–20

–30

–40–60 –40 –20 0 20 40 60 80 100

Days after transplanting

B


in the uncompacted C0 plots (6.7 wk). In that field, however, no floodwater wasobserved from 55 and 69 DAT onward for the C0 and C1 treatments, and therefore therice suffered from severe water stress during the second half of its reproductive cycle.Thus, in both seasons, the effects of subsoil compaction were limited to a few extracentimeters in the level of the perched water table (from –42 to –15 DAT) and depthof floodwater (from –15 to 53 DAT) in C1 compared with C0 plots (Fig. 5B). Thisresult is in contrast with earlier reports (Sharma et al 1995a,b, Trébuil et al 1998),emphasizing the importance of subsoil clay percentage, toposequence position, andgroundwater table depth on the response to subsoil compaction.

Soil moisture content. In the 1993 WS, volumetric soil moisture measurementsmade after the disappearance of standing water did not show any significant differ-ences between treatments. In the 1994 WS, the perched water table observed in theC1 compacted plots receded almost as rapidly as in the uncompacted plots at the endof the wet season (Fig. 5B). Similar observations were made in 1993 concerning thelack of differences between treatments for volumetric soil moisture content at the endof the wet season. Such a rapid depletion of the soil profile water content, because ofthe very sandy soil texture and high infiltration rates, even after subsoil compaction,shows the lack of effectiveness of subsoil compaction to conserve water for the ricecrop for use during the last month of its reproductive phase, although this may varywith the soil factors discussed above.

These observations show that the understanding of subsurface hydrology at thefield and landscape levels must be taken into account when delineating soil suitabilityfor subsoil compaction. Further, the variation in response between on-station and on-farm studies illustrates the importance of on-farm testing.

Effects of subsoil compaction and green manureon rainfed lowland rice productivityUnder the conditions of the 1993 WS, mean yields measured in the C1 and C2 com-pacted plots were 1.24 and 1.18 t ha–1, respectively, compared with 1.31 t ha–1 inuncompacted C0 plots. With standard errors of the yields of about 0.6 t ha–1, no sig-nificant differences in rice grain yields between either compaction or tillage depthtreatments were observed. In the 1994 WS, with a mean value of 1.3 t ha–1, grainyields obtained in compacted plots were 17% higher than in the C0 uncompactedplots. However, large standard errors of the yields led to a statistically nonsignificantdifference in final grain yields between the main plots (Fig. 6). Such large standarderrors reflect substantial variation from field to field within each treatment, and withno replication within fields.

Green manure before rice generally increased grain yields by only 8% and itseffect in uncompacted plots was even more limited. On average, the combination ofsubsoil compaction and green manure in C1F1 plots increased grain yields by 25% incomparison with the C0F0 treatments, but this difference was also not statisticallysignificant. In fields with at least 2% subsoil clay content, statistically nonsignificantdifferences in dry matter production at heading were observed, which reflected thepositive effects of compaction and green manuring on rice growth (data not shown).


Delineating soil suitability for subsoil compactionin the Ubon Ratchathani Land Reform AreaDelineation of soil domains for which subsoil compaction could be suitable wasachieved by reclassifying the map of clay content in the subsoil. Clay content wascategorized using thresholds of 2%, 5%, and 10% clay in the subsoil. Figure 7 identi-fies the 250 × 250-m cells that have 2–5% and 5–10% of clay in the subsoil on 40,000ha within the Ubon Ratchatani Land Reform Area. Approximately 40% of the mappedarea has clay content between 2% and 10%. This land may be suitable for subsoilcompaction. It should be noted that subsoil compaction in this study was based on theclay content of the 30–60-cm layer, whereas the map in Figure 7 was produced fromthe available data in Oberthür et al (1999), where the subsoil was defined as the 15–40-cm layer. Figure 3 shows, however, that soil penetration resistance within 15–60-cm depth did not change as much as within 0–15-cm depth. This may be an indicationthat soil texture did not change very much in the 15–30-cm layer. Figure 3 also showsthat the layer between 15- and 65-cm depth was most affected by the compaction.Data for the 15–30-cm layer were thus important in defining soil suitability for sub-soil compaction. The mismatch between the requirement for subsoil compaction andthe data available from the characterization work illustrates a common problem inidentifying suitability classes. There is a need to validate these findings before pro-ceeding further.

Fig. 6. Effects of subsoil compaction and green manure on rice grainyields in six farmers’ fields of Det Udom District, Ubon RatchathaniProvince, 1994 wet season. Fertilizer 0 = no green manure, fertil-izer 1 = with green manure.

3.0

2.5

1.4 2.1 2.3 2.6 4.4 4.6

Subsoil clay content (%)

Yield (t ha–1)

Uncompacted + fertilizer 0Uncompacted + fertilizer 1Compacted + fertilizer 0Compacted + fertilizer 1

2.0

1.5

1.0

0.5

0


Conclusions

Variations in ponded water depth and rice yield in response to subsoil compactionwere attributed to differences in subsoil texture and toposequence position, whichaffected the groundwater depth of the experimental fields. A minimum subsoil claycontent of 2% seemed to be necessary to ensure successful subsoil compaction of thecoarse-textured soils of the Nam Pong series. Even in such field situations, the infil-tration rates (which were largely affected by the observed hydraulic conductivity ofthe subsoil) in compacted plots were still relatively high; therefore, subsoil compac-tion would not be expected to impede the production of annual crops such as grainlegumes. These findings need further confirmation.

The study illustrated a methodology using GIS and available secondary data todelineate areas with appropriate subsoil clay contents for subsoil compaction. Care-ful interpretation of these areas is necessary, however, as the chosen approach toextrapolation did not address several characteristics that might prevent the successfuladoption of the technique. One important factor is the characteristic of the soil pro-files. On-farm compaction trials were located in areas with deep sandy horizons, al-though much of the mapped area has compacted illuvial horizons in the 0.3–0.8-msoil depth. Gravelly layers of hardened iron oxide pebbles are also frequently foundjust below the topsoil; these might prevent successful compaction. Further on-farm

Fig. 7. Suitability for subsoil compaction in a subregion of the Ubon RatchathaniLand Reform Area. Subsoil compaction may be effective in reducing percolationin areas with clay content between 2% and 10%.

Clay classified

≤ 2% 50 ha (<1%)

>2% but ≤ 5% 762 ha (2%)

>5% but ≤ 10% 1,365 ha (37%)

>10% 22,019 ha (60%)


experiments are required to investigate the effects of these environmental phenom-ena.

The study also illustrated that, apart from subsoil clay content, water table depthalso determined the effectiveness of soil compaction technology. To improve the presentdelineation of this soil suitability, data on groundwater table changes should be inte-grated. Unfortunately, this type of information is rarely readily available among sec-ondary data. Therefore, more carefully planned on-farm experiments taking into ac-count soil texture heterogeneities (particularly the percentage of subsoil clay contentand the clay type), and also groundwater table fluctuation during the wet season at themicro-catchment level, need to be carried out to delineate more precisely the soilsuitability for these farming practices, as well as to better understand interfield vari-ability and probability of response. Further investigations on these relatively costlytechniques are also needed to assess their economic profitability and social accept-ability in such a highly heterogeneous and variable rice ecosystem populated by agreat majority of resource-poor farmers.

ReferencesGarrity D, Vejpas C, Herrera W. 1992. Percolation barriers increase and stabilize rainfed low-

land rice yields on well drained soils. In: Murty VVN, Koga K, editors. Proceedings ofthe International Workshop on Soil and Water Engineering for Paddy Field Manage-ment, 28-30 Jan. 1992. Rangsit (Thailand): Asian Institute of Technology. p 413-421.

Goovaerts P. 1997. Geostatistics for natural resources evaluation. New York (USA): OxfordUniversity Press.

Oberthür T, Goovaerts P, Dobermann A. 1999. Mapping soil texture classes using field textur-ing, particle size distribution and local knowledge by both conventional and geostatisticalmethods. Eur. J. Soil Sci. 50:459-479.

Sharma PK. 1992. In situ water conservation in sandy soils for rainfed lowland rice. Rice Res.J. Bangkok (Thailand) 1:59-64.

Sharma PK, Ingram KT, Harnpichitvitaya D, De Datta SK. 1995a. Management of coarse-textured soils for water conservation in rainfed lowland rice. In: Ingram KT, editor.Rainfed lowland rice: agricultural research for high-risk environments. Manila (Philip-pines): International Rice Research Institute. p 167-177.

Sharma PK, Ingram KT, Harnpichitvitaya D. 1995b. Subsoil compaction to improve water useefficiency and yields of rainfed lowland rice in coarse-textured soils. Soil Till. Res.36:33-44.

Trébuil G, Harnpichitvitaya D, Tuong TP, Pantuwan G, Wade LJ, Wonprasaid S. 1998. Im-proved water conservation and nutrient-use efficiency via subsoil compaction and min-eral fertilization. In: Ladha JK, Wade LJ, Dobermann A, Reichardt W, Kirk GJD, PigginC, editors. Rainfed lowland rice: advances in nutrient management research. Proceed-ings of the International Workshop on Nutrient Research in Rainfed Lowlands, 12-15October 1998, Ubon Ratchathani, Thailand. Manila (Philippines): International RiceResearch Institute. p 245-256.


NotesAuthors’ addresses: D. Harnpichitvitaya, G. Pantuwan, and D. Suriya-Arunroj, Rice Research

Center, Ubon Ratchathani, Thailand; G. Trébuil, Department of Annual Crops, Centrede coopération internationale en recherche agronomique pour le développement (CIRAD-CA), BP 5035, 34032, Montpellier Cedex 1, France. Seconded to Agronomy, Plant Physi-ology, and Agroecology Division, International Rice Research Institute, Los Baños,Laguna, Philippines; T. Oberthür, International Center for Tropical Agriculture, Cali,Colombia; I. Craig, Land Reform Area Development-AIDAB project for Det UdomDistrict, Ubon Ratchathani Province; T.P. Tuong and L.J. Wade, International Rice Re-search Institute.


Modeling water availability, crop growth, and yield . . . 111

Characterizing and understanding the nature of the limitations imposed byrainfed lowland environments within a target region are important for devel-oping an efficient rice breeding program. Major factors determining the envi-ronment as it influences plant performance are soil properties and wateravailability. Rainfall is a major determinant of the water environment, but itsseasonal variability is high in northeast Thailand. Simulation models are use-ful in determining many features of the paddy water environment, as theycan readily estimate water availability as a probability function by using pastrainfall patterns as inputs. The RLRice model was developed for simulatingthe paddy water balance and growth of cultivar KDML105 for the rainfedlowlands in Thailand.

The model was used to quantify the water balance of paddies of theseveral locations used in the rainfed lowland rice breeding program in north-east Thailand. Yearly variation in simulated yield at any location was relatedto variation in rainfall during crop growth. However, sensitivity analysis re-sults showed that simulated yield varied greatly with changes in componentsof the water balance, particularly the deep percolation rate, lateral watermovement, and initial water level at transplanting. Simulated yield was gen-erally associated with the time of disappearance of water relative to flower-ing, and the depth of the free water level at flowering. The simulation resultsindicate a strong interaction of genotype and water environment through varia-tion in water availability in relation to phenological development of each geno-type.

Progress in breeding for high-yielding cultivars for rainfed lowland rice is slow inmany countries, partly because genotypic ranking for yield varies greatly in differentenvironments as a result of large genotype-by-environment (G × E) interactions forgrain yield. Often, the G × E interaction variance for yield is several times greaterthan the genotype variance alone in rainfed lowland rice (Cooper et al 1999a). Thelarge interaction for yield is related to the large environmental variation within the

Modeling water availability,crop growth, and yieldof rainfed lowland ricegenotypes in northeast ThailandS. Fukai, J. Basnayake, and M. Cooper

112 Fukai et al

target area of a breeding program. Wade et al (1999b) have documented the spatialand temporal variability in growing environments and yield in rainfed lowlands withinAsia. Variation occurs because of soil type and biotic factors, but the main cause ofthe large yield variation is often related to variation in water availability in rainfedlowlands. Water availability in lowland fields affects the timing and intensity of droughtand submergence development, and also causes reduced nutrient availability (Fukaiet al 1999a). Thus, yield is often reduced with drought because plant growth is af-fected by reduced water and nutrient availability.

Simulation models have been used successfully to determine the target envi-ronments of a breeding program in some crops (e.g., sorghum, Chapman and Barreto1997). This area of work is being used to help develop new cultivars that are suitablefor target environments. The approach has not yet been applied to the characteriza-tion of environments for rainfed lowland rice, partly because of the difficulty of quan-tifying the water balance for rainfed lowlands, particularly lateral water movementacross paddies.

This chapter aims to evaluate factors that determine water availability in rainfedlowlands and demonstrate how a change in water balance components may interactwith genotypes of different phenology. The RLRice model (Fukai et al 1995) wasused to estimate the water balance and grain yield for a range of environments innortheast Thailand. The mean yield in northeast Thailand is 1.7 t ha–1 and popularcultivar KDML105 can produce 4.0 t ha–1 under favorable conditions.

Water balance in rainfed lowlands

The position of lowland fields in a toposequence is important in determining wateravailability (Wade et al 1999b). Lowland fields in the high or upper terraces lose alarge amount of water readily, particularly after heavy rainfalls, through surface run-off and underground lateral water movement, while those in the lower terraces mayintercept the water that flows in from the paddies in an upper position. This lateralwater movement results in different periods of water availability and growing dura-tion, often more than 30 days within a small area. Thus, the upper terraces may beclassified into the drought-prone subecosystem and the lower terraces may belong tothe submergence-prone or drought- and submergence-prone subecosystem. Most farmsare composed of a mixture of these different positions in the toposequence in varyingproportions, depending on their location. When selection trials are conducted in arainfed lowland rice breeding program, the positions of the trials in the toposequenceare likely to have large effects on the relative yield of the breeding lines and thereforeon the lines selected.

The change in soil water content or free water level above or below the soilsurface is described by the equation

∆S = R – (E + T + P + L + O)


where ∆S is the change in soil water content, R the rainfall, E the evaporation fromstanding water surface or soil surface, T the transpiration, P the deep percolation, Lthe net lateral water movement in the soil (positive means loss of water from theparticular field), and O the runoff of water above the bund (Fig. 1).

Water availability during growth has a direct effect on grain yield in rainfedlowland rice. Jearakongman et al (1995) found that yield was high when standingwater was maintained until flowering, but it decreased sharply when standing waterdisappeared before flowering. Because of late flowering, late-maturing cultivars weremore affected than early maturing cultivars by late-season drought. The interactionbetween environmental variation in the timing of drought and plant development hascontributed to many of the large G × E interactions for grain yield of rainfed lowlandrice that are observed in the multienvironment trials conducted in Thailand.

The deep percolation rate varies greatly, and this causes large variation in totalwater loss from lowland fields (Yahata 1976). One way of altering the water balanceis by puddling or compacting the soil to reduce the deep percolation rate and thisaffects grain yield (Sharma et al 1988). The effect of variation in deep percolation rateon grain yield, however, depends on the phenology type of the cultivar used (Fukai1996).

The RLRice model

A rainfed lowland rice model was developed for cultivar KDML105 (Fukai et al1995). A brief account of this model is given here, concentrating on aspects of themodel that are relevant to this study.

The model consists of several sections. Daily rainfall, mean temperature, solarradiation, and pan evaporation are weather inputs to the model. The water balancesubmodel uses the equation mentioned earlier to estimate daily water balance fromthe time of transplanting to maturity. Both evaporation (E) and transpiration (T) aredetermined from daily pan evaporation, soil water content, and canopy ground cover,

Rainfall Transpiration

RunoffEvaporation

LateralmovementPercolation

Fig. 1. Diagrammatic representation of the components involved in waterbalance in terraced rainfed lowland paddies.

114 Fukai et al

which is calculated from the estimated leaf area of the crop. The transpiration ratedecreases when soil water content is less than 75% of the total extractable soil watercontent. The deep percolation rate (P) is a characteristic of the soil type at a locationand is a required input for the model. The deep percolation rate is assumed to be 0when the free water surface is below the effective root zone.

The model has two soil layers: the top 10 cm and the subsurface soil layer. Soilevaporation takes place only from the top layer, whereas T can occur from both lay-ers. The water-holding capacity for each soil layer has to be estimated and is an input.Roots are assumed to extend at the rate of 2 cm d–1. Parameters E and T are reduced assoil water content decreases in the top and bottom soil layers, respectively. Lateralwater movement (L) depends on rainfall, soil water level, and the position of thelowland field in a toposequence. This is estimated from a coefficient CL and rainfall(R) such that L = R*(CL – 1) if the soil is saturated with water and rainfall is large.Thus, the total water available to the paddy from rainfall is estimated as R*CL; CL>1.0 indicates that the lowland field is located at the lower part of the toposequenceand gains water from net lateral water movement. Net loss of water by lateral watermovement is estimated as R*(1.0 – CL) where CL <1.0. When CL = 1.0, the lowlandfield is located in a position where there is no net lateral water movement. Floodwaterlevel at transplanting varies in each crop and is an input. The subsequent water levelbelow or above the soil surface is estimated from R, E, T, P, L, and O throughoutgrowth.

Crop growth rate (CGR) is calculated daily from incident solar radiation, theproportion of radiation intercepted by the crop canopy, and radiation-use efficiency(amount of biomass produced per unit radiation intercepted) when soil water does notlimit growth. Radiation-use efficiency is affected by soil fertility and is an input foreach location. The proportion of radiation intercepted by the crop canopy is estimatedfrom the leaf area index. The leaf area index is estimated from dry matter production,partitioning of assimilates to leaves, and the ratio of leaf area to leaf dry weight.When plants are water-stressed, CGR is calculated as the product of T and transpira-tion efficiency (amount of biomass produced per unit water transpired). Accurateestimation of CGR is vital, as it determines not only total biomass production and leafarea index but also the yield components directly in the model. Panicle number m–2 isproportional to CGR before panicle initiation, whereas spikelet number panicle–1 isproportional to CGR between panicle initiation and anthesis. The unfilled grain pro-portion and single grain weight are determined directly by CGR after anthesis. Theseyield components are, however, also affected by water stress, which occurs near andafter anthesis. Thus, the proportion of unfilled grain increases as severe water stressdevelops near anthesis; the proportion of unfilled grain increases linearly with thenumber of stress days around anthesis, with the regression coefficient indicating agenotype’s tolerance of or susceptibility to drought near anthesis. Under severe wa-ter-stress conditions, the plants may die prematurely. In this case, grain yield is esti-mated from dry matter produced after flowering and a proportion of assimilates pro-duced before anthesis, which are translocated to fill grains.


Seeding date and transplanting date are inputs. If old seedlings are used fortransplanting, the time from transplanting to anthesis is shortened and the potentialfor high grain yield will be reduced.

While the model was developed for cultivar KDML105, the phenology of thecultivar can be modified in the model. The date of flowering is assumed to depend onthe extent of photoperiod sensitivity, the length of the basic vegetative phase in whichthe plants do not respond to photoperiod, and the actual date of sowing. Flowering isdelayed if severe water stress develops during the vegetative and reproductive stagebefore flowering. On the other hand, water stress after flowering hastens senescence.The stage of phenological development determines assimilate partitioning among roots,stems, leaves, and panicles, including grain. The proportion of assimilates allocatedto the roots and leaves decreases gradually while that allocated to the stem increasesafter transplanting until about the time of anthesis. Late-flowering genotypes have alonger time for leaf development and, hence, given an adequate water supply, have ahigher grain yield potential.

The model was calibrated for rainfed lowland conditions using data collectedfor six different locations in Thailand. The simulated results have been comparedwith experimental results for KDML105 and give good agreement (Fukai et al 1995).The model was used to estimate the magnitude of drought effects in northeast Thai-land (Jongdee et al 1997), G × E interaction for flowering time and grain yield(Henderson et al 1996), and water and nutrient availability interaction (Fukai et al1999a,b). The present work estimates how genotypes may interact with variation invarious components of the water balance equation to demonstrate G × E interactioncaused by variation in water availability.

Simulation procedures

Five sets of simulation experiments were conducted.1. Yearly yield variation due to rainfall for a given conditionat transplantingYearly variation for yield in relation to rainfall during the growth period was esti-mated from simulations for Ubon Ratchathani using 22 years’ rainfall data from 1975to 1996. For these simulations, all input values other than rainfall were constant foreach year and were those obtained in an experiment conducted in 1993. The standardinputs used for Ubon are as follows: time of seeding, 16 July; age of seedlings attransplanting, 40 d; standing water level at transplanting, 10 mm; lateral water move-ment coefficient, 1.0; deep percolation rate, 6.0 mm d–1 (Jongdee et al 1997).

The time when standing water disappeared from the lowland field was esti-mated for each year and this time was expressed in relation to flowering time. Freewater level at flowering was also used as an indicator of water availability to the crop.

2. Sensitivity analyses of water balance componentsThe location of selection trials within a small area may result in different values of thewater balance components, and may have a large effect on water availability and

116 Fukai et al

hence grain yield. The three major parameters that affect the water balance in rainfedlowlands—standing water at transplanting, lateral water movement coefficient, anddeep percolation rate—are altered independently of each other to examine the magni-tude of the effect of variation in each parameter value (sensitivity analysis). In thesesimulation runs, the standard values of the parameters for Ubon, except for the onebeing examined, are used together with Ubon 1996 rainfall data. The simulation wasrepeated for Phrae using the standard parameter values for Phrae and the 1993 rainfalldata for the location. The standard values were time of seeding, 8 July; age of seed-lings at transplanting, 38 d; standing water level at transplanting, 50 mm; lateral wa-ter movement coefficient, 1.0; and deep percolation rate, 1.0 mm d–1.

The standing water at transplanting was either –200 or 50 mm. The formersimulates the case where there is no standing water at transplanting but the lowlandfield would be moist enough to allow successful transplanting of seedlings. The latteris considered to represent the maximum water depth at transplanting. Values for thecoefficient for lateral water movement (CL) used were 0.5, 0.75, 1.0, 1.25, and 1.5.The values of 0.5 and 0.75 would represent the fields at high positions in a toposequenceand 1.25 and 1.5 the fields at low positions. Deep percolation rates used were 1, 4,and 6 mm d–1. The common range of this parameter is 1–6 mm d–1 in northeast Thai-land (Fukai et al 1995).

3. Response of genotypes with different phenologyto variation in water balance componentsIn this section, five genotypes of different phenology type were used for simulationfor Ubon. These were the same as those used by Henderson et al (1996) for simula-tion of G × E interaction analysis and they cover the range of phenology types com-mon in northeast Thailand. The standard genotype was KDML105, which is photope-riod-sensitive and late-flowering, commonly flowering around 25 October in north-east Thailand. Chiangsaen is strongly photosensitive and later flowering thanKDML105. RD23 is almost completely photoperiod-insensitive and early flowering.NSG19 and IR57514-PMI-5-B-1-2 are both mildly sensitive, but NSG19 floweredearlier than IR57514-PMI-5-B-1-2. Only phenology was altered in these simulationsand other cultivar characteristics remained the same as for KDML105.

The simulations were run for the higher, middle, and lower positions of atoposequence by using the lateral water movement coefficient (CL) of 0.5, 1.0, 1.25(only for the 16 July simulation in 1994), and 1.5, respectively. The two seeding datesused were 16 June and 16 July using rainfall data for Ubon in both 1994 (develop-ment of some water stress) and 1996 (favorable water conditions). For the simulationwith CL of 1.25 in the 16 July seeding, the deep percolation rate was altered. The deeppercolation rate may be lower in lower toposequence positions as the soil tends tohave more clay content (Wade et al 1999b). This was simulated by changing the deeppercolation rate from the standard of 6 to 4 and 2 mm d–1. The standard value forwater-holding capacity of the soil for Ubon is 55 mm, and the effect of an increase to75 mm was simulated. Wade et al (1999b) showed variation of about 20 mm along thetoposequence in their measurements in Bangladesh.


4. Effects of seedling age at transplantingAnother source of G × E interaction is the use of different seedling ages at transplant-ing in different trials. A genotype’s response to seedling age at transplanting may bedifferent in different water availability conditions. The simulation was conducted forUbon using 1994 and 1996 rainfall data. Seedling ages used were 25, 40, and 55 dold. The 25-d-old seedling would represent young seedlings used when water is avail-able at the optimum transplanting time. The older seedlings represent cases wherewater is not sufficient in lowland fields when seedlings are young and transplanting isdelayed for 15–30 d.

5. Yield and yield stability of genotypes with different phenologyThis set of simulations used the same five genotypes with different phenology men-tioned above and the standard parameter values obtained from experiments conductedin 1993 at nine locations in Thailand: three from northern Thailand and six fromnortheast Thailand (Fig. 2). The simulation was conducted using the past 20 years’

Fig. 2. Map of Thailand showing nine northern and northeastern loca-tions where cultivar comparisons were made.

SPT

PRE

PSL CPA KKN

PMISRN

SKN

UBN

Myanmar

Lao PDRVietnam

Gulf ofTonkin

Cambodia

Gulf ofThailand

Malaysia

Andaman Sea

Bangkok

118 Fukai et al

rainfall data for each location. Mean yield and standard deviation of yield were esti-mated for each genotype at each location.

Results

1. Yearly yield variation due to rainfall for a given conditionat transplantingThe simulated yield level at Ubon was generally low in these simulations becauseseeding occurred rather late (16 July), the deep percolation rate was high (6 mm d–1),and soil fertility was low.

Figure 3 shows water levels in the field from transplanting to maturity for fourcontrasting years; 1980 was one of the highest rainfall years, with standing waterlasting until flowering, whereas 1986 had periods in early vegetative stages whenthere was no standing water and standing water disappeared rather early, resulting inyield of 0.93 t ha–1. The third year, 1993, was a very dry year and standing waterdisappeared well before flowering, resulting in a low yield of 0.58 t ha–1. In 1996, thewater level was low immediately after transplanting and standing water also disap-peared about 10 d before flowering.

Yearly variation in yield was mostly accounted for by the variation in rainfallduring the period from transplanting to maturity (Fig. 4A). In years of high rainfall,

Fig. 3. Simulated water levels during growth and flowering of genotype KDML105 forUbon Ratchathani in four selected years. The letter “F” indicates time of flowering.

Days after sowing

400

0

–400

–800

–1,200

Water level (mm)

F

Simulated yield = 1.94 t ha–1

1980

F


1986

F

40 60 80 100 120 140


1996

400

0

–400

–800

–1,200

F

40 60 80 100 120 140



Fig. 4. Relationship between simulated grain yield and (A) rainfall from trans-planting to maturity, (B) date of disappearance of standing water relative toflowering (– indicates before flowering and + indicates after flowering), and (C)water levels at flowering for 22 years for Ubon Ratchathani.

Grain yield (t ha–1)

9588

81

83

829680

84

87

79

9377

867885

89 94

929091 76

75

0 100 200 300 400 500 600 700 800

Rainfall (mm)

R2 = 0.67Y = –0.49 + 0.0037X

A

3

2

1

0

R2 = 0.61Y = 1.74 + 0.0018X

C

3

2

1

0–1,000

Water level at flowering (mm)

8193 95

77

79

7892 90

83

8296 75 80 84

8776

9186

858994

88

–800 –600 –400 –200 0 200 400

R2 = 0.42Y = 1.47 + 0.020X

B

3

2

1

0

9588

8193

8577

79

9286788994

7683

8296

87

75

9091

8084

–60 –40 –20 0 20

Time of disappearance of standing water(days to flowering)

F

120 Fukai et al

such as 1980 and 1996, yield was high, whereas, in years of low rainfall, such as1993, yield was low. In years of high rainfall, standing water disappeared from thelowland field later and the water level at flowering was higher (Fig. 4B,C). Yield wasbelow 1 t ha–1 when standing water disappeared more than 20 d before flowering orthe water level at flowering was less than 400 mm below the soil surface.

2. Sensitivity analyses of water balance componentsThe standing water level at transplanting showed a large simulated effect on grainyield because water availability to the crop during subsequent growth periods wasmodified. In Figure 5, the simulated water level throughout growth is shown whenthe initial water level at transplanting was –200 mm and 50 mm. With the initial waterlevel at 200 mm below the soil surface, there was no standing water for 15 d aftertransplanting and standing water also disappeared several days earlier. This resultedin a simulated yield of 1.41 t ha–1 versus 1.84 t ha–1 for the 50-mm water treatmentwith a longer period of standing water. Figure 6 shows simulation results of the effectof a change in the lateral water movement coefficient (CL) in 1996 together with theresults in 1994. The period of standing water increased markedly with the increase inCL from 0.5 to 1.5 in both years.

Fig. 5. Simulated water levels during the growth of genotypeKDML105 for two levels of standing water at the time of trans-planting at Ubon Ratchathani. The letter “F” indicates time of flow-ering.

Days after sowing

40 60 80 100 120 140

Simulated grain yield = 1.84 t ha–1

50 mm (water level at transplanting)

F

0

–400

–800

–1,200

0

–400

–800

–1,200

Water level (mm)

Simulated grain yield = 1.41 t ha–1

–200 mm (water level at transplanting)

F


Days after sowing

UBN 96 (0.5)

(E) (D)(B,C)

(A)

UBN 96 (1.0)

(E) (D)(B,C)

(A)

40 60 80 100 120 140

UBN 96 (1.5)(E) (D) (B,C)

(A)

400

0

–400

–80040 60 80 100 120 140

UBN 94 (1.5)

(E) (D) (B,C) (A)

400

0

–400

–800

–1,200

UBN 94 (1.0)

(E) (D) (B,C)(A)

400

0

–400

–800

–1,200

UBN 94 (0.5)

(E) (D) (B,C)(A)

Water level (mm)

Fig. 6. Simulated water levels under three levels of lateral water movement coefficients (0.5,1.0, and 1.5) during the growth of five genotypes: Chiangsaen (A), KDML105 (B), IR57514 (C),RD23 (D), and NSG19 (E) tested at Ubon Ratchathani in 1994 and 1996. Down arrows indicatetime of flowering of each genotype. The filled bars indicate the period when KDML105 had diedbecause of severe drought.

122 Fukai et al

Table 1 shows the results of the sensitivity analyses for simulated change inyield for the coefficient for lateral water movement and water level at transplantingand for the coefficient for lateral water movement and deep percolation rate. Thesimulation results show that the effect of a change in lateral water movement wouldbe greater with an initial water level of +50 mm than with –200 mm at Ubon 1996because lateral water movement would be greater with a longer period of standingwater and, hence, the soil is saturated with water for a longer period. When the waterlevel at transplanting was reduced to –200 mm, the period of standing water wasshorter, and this resulted in a rather small effect on grain yield with the variation inCL. At Ubon, the effect of a change in CL became smaller as the deep percolation ratedecreased from 6 to 1 mm d–1. With 1 mm d–1, there was almost no water stressthroughout the growth period and hence the effect of water movement was small,whereas, with 4–6 mm d–1, there were some periods with standing water but therewere also periods of water stress (Fig. 5 for a deep percolation rate of 6 mm d–1). Inthis case, there was a large simulated effect of variation in lateral water movement.

The results are different for Phrae 1993, where yield potential was higher andsimulated yield was more responsive to a change in initial water level and deep per-

Table 1. Simulated grain yield (t ha–1) for KDML105 under differentdegrees of lateral movement of water and initial water level (mm) atthe time of transplanting and deep percolation rate (mm d–1) at Ubon(1996) and Phrae (1993).

Coefficient for lateral movementWater level at of water (CL)transplanting

0.5 0.75 1.0 1.25 1.50

Ubon–200 1.32 1.34 1.41 1.45 1.52+50 1.56 1.72 1.84 2.12 2.26

Phrae–200 0.94 1.41 2.25 2.97 3.18+50 2.08 2.89 3.26 3.49 3.69

Coefficient for lateral movementDeep percolation of water (CL)

rate0.5 0.75 1.0 1.25 1.50

Ubon1 2.28 2.46 2.46 2.46 2.464 1.78 1.85 2.15 2.37 2.416 1.56 1.73 1.79 1.88 2.12

Phrae1 2.08 2.89 3.26 3.49 3.694 0.37 0.61 0.89 1.34 1.816 0.35 0.49 0.51 0.76 0.88


colation rate. In this case, the larger effect of variation in lateral water movement wasfound with lower water availability (–200 mm) at transplanting and a lower percola-tion rate. With the highest percolation rate of 6 mm d–1, the water-stress effect wassevere and the effect of variation in CL was small.

In these simulations, grain yield was directly related to the date of disappear-ance of standing water relative to the flowering date for both locations and Figure 7shows the results for Ubon. Note that relationships obtained from the sensitivity analy-ses are similar to those obtained from the analysis of yearly variation (Fig. 4B,C).

3. Response of genotypes with different phenologyto variation in water balance componentsFor the June planting in 1994 at Ubon, the yield of genotypes with different phenol-ogy increased with a delay in flowering until about 10 October with the lateral watermovement coefficient of 0.5 and until about 20 October with the coefficients of 1.0and 1.5 (Fig. 8). However, with seeding later in July, earlier flowering was advanta-geous with a CL of 0.5, 1.0, and 1.25, whereas the phenology had a small effect whenthe coefficient was 1.5, with a slight advantage with later flowering. It should bepointed out that the order of flowering among genotypes changed between the twoseeding dates because of differences in photoperiod sensitivity. For example, the mildlyphotosensitive IR57514-PMI-5-B-1-2 flowered earlier than KDML105 when plantedon 16 June, but they flowered at almost the same time when sown on 16 July.

In 1996, the water conditions were generally more favorable and the highestyield was obtained with the KDML105 phenology type for all values of CL in bothJune and July sowing.

In Figure 6, where the water level throughout the growth period for the Julysowing is shown, flowering time of each genotype is also shown for both 1994 and1996. With a CL of 0.5 and 1.0, the three late genotypes flowered well after the stand-ing water disappeared from the lowland field and the yields were lower. With the CLof 1.5, standing water was maintained until the last genotype flowered and the later

Fig. 7. Relationship between the simulated grain yield and date of disappearance of waterrelative to flowering time for five genotypes. (A) with two levels of standing water at transplant-ing and (B) three levels of deep percolation rate at Ubon Ratchathani in 1996.

3

2

1

0

–30 –20 –10 0 10 20 30


F

AR2 = 0.73Y = 1.89 + 0.025X

50 mm–200 mm

3

2

1

0–30 –20 –10 0 10 20 30

F

B

R2 = 0.97Y = 1.99 + 0.018X

6 mm d–1

4 mm d–1

1 mm d–1

Time of disappearance of standing water(days to flowering)

124 Fukai et al

Fig. 8. Simulated grain yield in relation to flowering date for fivegenotypes seeded at two different times in 1994 and 1996 (June,open symbols; July, closed symbols) under four levels of lateralwater movement coefficients ( 1.5; 1.25; 1.0; 0.5 mm d–1) at Ubon Ratchathani (UBN).

genotypes were not disadvantaged. In 1996, water disappearance was several dayslater, and this resulted in the later flowering genotypes maintaining yields compa-rable with those of the early genotypes, even with the CL of 0.5.

The simulation results of a change in deep percolation rate for Ubon 1994 wherethe CL of 1.25 was used suggest that later flowering genotypes may not be disadvan-taged at the lower toposequence position if the deep percolation rate was reducedfrom the standard 6 mm d–1 to 2–4 mm d–1 (Fig. 9). The effect of a change in soilwater-holding capacity from 55 to 75 mm was small, however. Yield increased from

Flowering date

13 Sep

UBN 1996

23 Sep 3 Oct 13 Oct 23 Oct 2 Nov 12 Nov

3

2

1

0

RD23

NSG19

IR57514

KDML105

CS

3

2

1

0


UBN 1994

.


1.0 to 1.3 t ha–1 with a deep percolation rate of 6 mm d–1, but the increase was smallerwith a deep percolation rate of 4 mm d–1.

4. Effects of seedling age at transplantingThere were large adverse effects for the use of old seedlings for transplanting, and theeffects were greater in 1994, when stress developed earlier (Table 2). Crop growth

Table 2. Simulated effects of seedling age attransplanting on grain yield (t ha–1) for five ricephenology groups (genotypes) grown at Ubon;1994 and 1996 rainfall data were used to simu-late the grain yield of these genotypes.

Seedling age at transplantingGenotypes

25 d 40 d 55 d

1994NSG19 1.96 1.17 0.55RD23 2.02 0.96 0.48IR57514 1.79 0.52 0.17KDML105 1.88 0.54 0.19Chiangsaen 1.38 0.39 0.09

1996NSG19 1.79 1.55 1.10RD23 1.83 1.61 1.15IR57514 1.93 1.79 1.41KDML105 1.92 1.80 1.39Chiangsaen 1.60 1.43 1.09

Fig. 9. Simulated grain yield in relation to flow-ering date of five genotypes tested under a lat-eral water movement coefficient of 1.25 mmd–1 and three levels of deep percolation rate atUbon Ratchathani in 1994.

3

2

1

0


23 Oct 2 Nov 12 NovFlowering date

RD23NSG19IR57514KDML105CS

2 mm

4 mm

6 mm

13 Oct

126 Fukai et al

Table 3. Simulated grain yield (t ha–1) (average of 20 years) and their standarddeviation (Std) for five phenology groups grown in nine locations in Thailand.

Cultivar

Loca- NSG19 RD23 IR57514 KDML105 Chiangsaentiona

Mean Std Mean Std Mean Std Mean Std Mean Std

CPA 1.82 ±0.68 1.84 ±0.68 2.64 ±1.08 1.86 ±0.74 2.79 ±1.39KKN 1.74 ±0.24 1.53 ±0.21 2.13 ±0.29 3.23 ±0.25 2.33 ±0.41PMI 2.50 ±1.29 2.21 ±0.85 2.08 ±0.78 1.64 ±0.79 2.22 ±1.28PRE 2.36 ±0.36 2.29 ±0.47 3.03 ±1.06 3.02 ±0.47 3.30 ±1.38PSL 2.29 ±0.27 2.49 ±0.37 3.27 ±0.75 3.23 ±0.74 3.33 ±1.01SKN 2.36 ±0.58 2.14 ±0.41 2.43 ±0.68 2.32 ±0.80 2.01 ±0.84SPT 2.19 ±0.64 2.39 ±0.78 2.67 ±1.06 2.73 ±1.06 2.76 ±1.19SRN 3.26 ±0.35 2.85 ±0.34 4.01 ±0.33 4.33 ±0.35 4.62 ±0.40UBN 1.33 ±0.31 1.26 ±0.39 1.01 ±0.50 1.03 ±0.53 1.04 ±0.56

aCPA = Chum Phae, KKN = Khon Kaen, PMI = Phimai, PRE = Phrae, PSL = Phitsanulok, SKN =Sakon Nakhon, SPT = Sanpatong, SRN = Surin, UBN = Ubon Ratchathani.

decreased with the use of old seedlings because the period from transplanting to flow-ering in the lowland field was shortened. This was particularly obvious when waterstress developed. However, interaction with genotypes was relatively small; it wasapparent only in 1994, when early flowering genotypes (NSG19, RD23) were advan-tageous under delayed transplanting. As the sowing was late at 16 July, IR57514-PMI-5-B-1-2 (mildly photosensitive) and KDML105 (strongly photosensitive) hadsimulated flowering at about the same time, and the effect of delayed transplantingwas almost the same in both years.

5. Yield and yield stability of genotypes with different phenologyThere were genotype-by-location interactions for mean yield in these simulations(Table 3). The highest yield was obtained with later maturing genotypes (KDML105,Chiangsaen) at Surin and Phrae, where yield was generally high, with little adverseeffect of late-season drought. The earlier genotypes (NSG19, RD23) produced a higheryield than the others at Phimai and Ubon, where late-season drought was a commonproblem. These interactions reflect rainfall patterns and the standard water balanceparameter values used for the different locations. On the other hand, the standarddeviation for yield was commonly higher for later-flowering genotypes. These resultsindicate the more stable nature of early flowering genotypes compared with late-flowering genotypes. The standard deviation of each genotype differs markedly amonglocations. Yield obtained at Phimai generally had a larger standard deviation than atthe other locations, whereas it was small at Khon Kaen and Surin.


Discussion

This simulation work shows the importance of maintaining standing water for a longtime period during the reproductive to grain-filling stages. This confirms the experi-mental results in northeast Thailand reported by Jearakongman et al (1995). Twoindicators can be used to estimate overall water availability and grain yield in a plantbreeding program: time of disappearance of standing water relative to flowering timeand free water level at flowering. The latter can be more readily determined if there isstanding water at flowering, but this may not be the case if standing water has alreadydisappeared by flowering. Recording the date of disappearance of standing water anddates of flowering of different genotypes should help characterize the water condi-tions of each selection trial.

The water balance in rainfed lowlands is complex and the time of disappear-ance of standing water relative to flowering is determined by several factors, as dem-onstrated in the simulation studies here. For a given hydrological and agronomic con-dition, rainfall between transplanting and flowering affects the relative time of disap-pearance of standing water. However, deep percolation rate, lateral water movement,and, to a lesser extent, initial water level at transplanting all affect water balance andhence grain yield. Sensitivity analyses of the influence of these factors on grain yield(Table 1) show interactions among these factors as well as with rainfall. These com-plex relationships would affect the water balance of a lowland field used by a breed-ing program. Among these interacting factors, rainfall can be determined accuratelyand initial water level can be readily estimated if there is standing water at transplant-ing. The deep percolation rate and lateral water movement are important componentsin determining overall water balance, and a combined rate can be determined readilyby monitoring the field water level and using the estimated value of the evapotranspi-ration rate. A separate estimation of the deep percolation rate and lateral water move-ment would make a more accurate estimation of water balance possible. This may beparticularly important in a toposequence where the lateral water movement compo-nent is expected to be large.

Lateral water movement is a difficult component of the water balance to deter-mine because it is not constant throughout the growth period. The model estimatesthe net amount of lateral water movement from the coefficient CL, the amount ofrainfall, and soil water level, thus assuming that lateral water movement takes placeimmediately after a large rainfall event and only when the soil is saturated with water.The value of the coefficient may range from 0.5 to 1.5 as used in the sensitivityanalysis in the present work, as this would cause about 30 d of difference in the dateof disappearance of standing water in a toposequence. This difference is not uncom-mon in rainfed lowlands (Wade et al 1999b) and cultivars with different maturitygroups are often planted at different toposequence positions (Nesbitt and Phaloeun1997). The development of a method to determine lateral water movement appearsimportant for the characterization of rainfed lowland environments. Water loss throughunder-bund percolation may also be important in determining the water balance, andmay need to be considered in the model (Tuong et al 1994).

128 Fukai et al

The use of five genotypes with different phenology, some of which correspondto the proposed reference lines for rainfed lowland rice (Wade et al 1999a), shows theG × E interaction due to genotypic phenology differences and water availability, par-ticularly during later growth stages. These interactions are mostly explained by thetime of disappearance of standing water relative to flowering of each genotype, thusemphasizing the importance of the water balance in each screening field in determin-ing the ranking of genotypes for yield. The genotypes used for the simulation wereassumed to be different only for phenology and not for other characters. Therefore,other important characters, such as potential yield under nonlimiting conditions,drought resistance, submergence tolerance, and tolerance for low nutrient availabilityand insects and diseases, are not included in the simulation. These factors are knownto affect genotypic performance (e.g., Fukai et al 1999a,b, Mackill et al 1999, Ito et al1999) in rainfed lowland rice, and hence further contribute to the G × E interactionsobserved in multienvironment trials.

In the model, genotypes with a later phenology are assumed to have a longertime for vegetative growth and hence will have a larger biomass at maturity, particu-larly under favorable water conditions. They generally resulted in higher yield whenthere were no periods of severe water stress (e.g., Fig. 6, sowing on 16 June and CL of1.0 or 1.5). This may not be the case in reality, as often later-flowering cultivars havea lower harvest index (Jearakongman et al 1995). Similarly, the potential yield ofphotoperiod-insensitive cultivars is often higher than that of strongly photoperiod-sensitive cultivars (Mackill et al 1996). These features are not simulated in the presentwork, although the RLRice model has the capacity to simulate modifications of thevariation in potential yield. The present work aimed to identify environmental fac-tors, particularly those associated with water availability, that cause G × E interac-tions for grain yield. The effects of variation in attributes of genotypes other thanphenology, such as drought resistance and potential yield, are the subject of anotherstudy.

While the initial water level at transplanting is an input to the model, this valueis affected by prior rainfall events as well as land preparation. Rainfed lowland breed-ers often irrigate the field for transplanting, and this practice ensures the use of youngseedlings for transplanting. This may, however, affect the selection of genotypes asshown in the simulation results in this study, which favor the selection of later-flow-ering genotypes. The simulated effect of the use of old seedlings on grain yield wassimilar to that obtained in the field in our recent experiments in Lao PDR.

This simulation study indicated the complex nature of the water balance inrainfed lowlands and suggested how this may interact with the phenology of geno-types in plant breeding experiments. Cooper et al (1999a,b) attributed much of thelarge genotype-by-site-by-year interactions for grain yield to the effects of environ-mental variation in the timing and intensity of water deficit relative to the floweringtimes of the genotypes. Given the strong influence of the water balance component ofthe rainfed lowland paddy on the relative yield performance of genotypes demon-strated in this study, it is important to consider the characterization of rainfed lowlandrice environments sampled at all stages of the plant breeding program if the compli-


cations of G × E interactions for yield are to be addressed. Further field studies arerequired to develop a method for accurately estimating lateral water movement forbetter characterization of rainfed lowland environments.

ReferencesChapman SC, Barreto HJ. 1997. Using simulation models and spatial database to improve the

efficiency of plant breeding programs. In: Cooper M, Hammer GL, editors. Plant adap-tation and crop improvement. Wallingford (UK): CAB International and InternationalRice Research Institute. p 563-590.

Cooper M, Rajatasereekul S, Immark S, Fukai S, Basnayake J. 1999a. Rainfed lowland ricebreeding strategies for northeast Thailand. 1. Genotypic variation and genotype × envi-ronment interactions for grain yield. Field Crops Res. 64:131-151.

Cooper M, Rajatasereekul S, Somrith B, Sriwisut S, Immark S, Boonwite C, Suwanwongse A,Ruangsook S, Hanviriyapant P, Romyen P, Porn-uraisanit P, Skulkhu S, Fukai S,Basnayake J, Podlich DW. 1999b. Rainfed lowland rice breeding strategies for northeastThailand. 2. Comparison of intrastation and interstation selection. Field Crops Res.64:153-176.

Fukai S. 1996. Crop physiological approaches to understanding rice production under waterlimited conditions. In: Crop research in Asia: achievements and perspectives. Proceed-ings of the 2nd Asian Crop Science Conference, 21-23 August 1996, Fukui, Japan.p 240-245.

Fukai S, Rajatasereekul S, Boonjung H, Skulkhu E. 1995. Simulation modeling to quantify theeffect of drought for rainfed lowland rice in Northeast Thailand. In: Fragile lives infragile ecosystems. Proceedings of the International Rice Research Conference. Manila(Philippines): International Rice Research Institute. p 657-674.

Fukai S, Inthapanya P, Blamey FPC, Khunthasuvon S. 1999a. Genotypic variation in rice grownin low fertile soils and drought-prone, rainfed lowland environments. Field Crops Res.64:121-130.

Fukai S, Pantuwan G, Jongdee B, Cooper M. 1999b. Screening for drought resistance in rainfedlowland rice. Field Crops Res. 64:61-74.

Henderson SA, Fukai S, Jongdee B, Cooper M. 1996. Comparing simulation and experimentalapproaches to analysing genotype by environment interactions for yield in rainfed low-land rice. In: Cooper M, Hammer GL, editors. Plant adaptation and crop improvement.Wallingford (UK): CAB International and International Rice Research Institute. p 443-486.

Ito O, Ella E, Kawano N. 1999. Physiological basis of submergence tolerance in rainfed low-land rice ecosystem. Field Crops Res. 64:75-90.

Jearakongman S, Rajatasereekul S, Naklang K, Romyen P, Fukai S, Skulkhu E, Jumpaket B,Nathabutr K. 1995. Growth and grain yield of contrasting rice cultivars grown underdifferent conditions of water availability. Field Crops Res. 44:139-150.

Jongdee S, Mitchell JH, Fukai S. 1997. Modelling approach for estimation of rice yield reduc-tion due to drought in Thailand. In: Fukai S, Cooper M, Salisbury J, editors. Breedingstrategies for rainfed lowland rice in drought-prone environments. Proceedings of anInternational Workshop held at Ubon Ratchathani, Thailand, 5-8 November 1996. p 65-73.

130 Fukai et al

Mackill DJ, Coffman WR, Garrity DP. 1996. Rainfed lowland rice improvement. Manila (Philip-pines): International Rice Research Institute. 242 p.

Mackill DJ, Nguyen HT, Jingxian Zhang. 1999. Use of molecular markers in plant improve-ment programs for rainfed lowland rice. Field Crops Res. 64:177-185.

Nesbitt HJ, Phaloeun C. 1997. Rice-based farming systems. In: Rice production in Cambodia.Manila (Philippines): International Rice Research Institute. p 31-37.

Sharma PK, Datta SK, Redulla CA. 1988. Tillage effects on soil physical properties and wet-land rice yield. Agron. J. 80:34-39.

Tuong TP, Wopereis MCS, Marquez JA, Kropff MJ. 1994. Mechanisms and control of percola-tion losses in irrigated puddled rice fields. J. Soil Sci. Soc. Am. 58(6):1794-1803.

Wade LJ, McLaren CG, Quintana L, Harnpichitvitaya D, Rajatasereekul S, Sarawgi AK, KumarA, Ahmed HU, Sarwoto, Singh AK, Rodriguez R, Siopongco J, Sarkarung S. 1999a.Genotype by environment interactions across diverse rainfed lowland rice environments.Field Crops Res. 64:35-50.

Wade LJ, Fukai S, Samson BK, Ali A, Mazid MA.1999b. Rainfed lowland rice: physicalenvironment and cultivar requirements. Field Crops Res. 64:3-12.

Yahata T. 1976. Physical properties of paddy soils in relation to their fertility. In: The fertilityof paddy soils and fertilizer applications for rice. Taipei (Taiwan): Food and FertiliserTechnology Centre for Asian and Pacific Region. p 28-35.

NotesAuthors’ address: School of Land and Food Sciences, The University of Queensland, Brisbane,

Australia.Acknowledgments: Financial support from the Australian Centre for International Agricultural

Research is gratefully acknowledged.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-


Using reference lines to classify multienvironment trials . . . 131

Using reference lines to classifymultienvironment trials to the targetpopulation of environments, and theirpotential role in environmentalcharacterizationC.G. McLaren and L.J. Wade

In heterogeneous rainfed environments, cultivar performance interacts withcrop management, soil type, topography, and agrohydrology to complicatethe task of selecting better-adapted cultivars. To make consistent progresswith selection in the presence of these genotype by environment interactions(G × E), it is important to clearly identify the target population of environ-ments and to know how well actual test environments represent this popula-tion. This chapter evaluates a methodology for using measurements on a setof reference lines to classify sites according to previously identified responsepatterns for a target population of environments. Strategies for choosingreference lines, classifying new sites, and deducing their environmental char-acteristics are examined.

The results showed that the reference set was able to capture repeat-able G × E patterns provided it contained representatives of all discrimina-tory genotype groups. The methodology for characterizing new environmentson the basis of reference line responses relied heavily on an ability to imputemissing values. Although no optimal solution was available, a heuristic solu-tion in the pattern analysis algorithm was satisfactory. Reference lines shouldbe chosen based on how well they match the discriminatory pattern of thegenotype group, their agronomic features, knowledge of their physiologicalresponses, and practical issues such as the availability of seed.

Based on this analysis, we conclude that a series of small field trials atmany locations could be employed to obtain a useful characterization of newenvironments and allow breeders to weight responses of test lines appropri-ately. If detailed physical and climatic measurements are also made in theseenvironments, the responses can be integrated with geographical informa-tion, physiological understanding, and crop modeling to quantify environmentfrequency, predictability, repeatability, and risk.

132 McLaren and Wade

In rainfed agriculture, crop performance is strongly influenced by climatic character-istics and by spatial heterogeneity over soil types, topographic sequences, andagrohydrologic conditions. Cultivar and management interact with these environmentalvariables, thus complicating the task of identifying improved cultivars. These inter-actions and the complexity of factors involved make it difficult to adequately definethe target population of environments or even to reliably assess cultivar performanceover those environments (Wade et al 1995, Cooper 1999). To provide a focus forselection programs, it is essential to clearly identify the target population of environ-ments, their spatial and temporal frequencies, and their characteristics.

Wade et al (1996) discussed two broad approaches for characterization of envi-ronments: one based on analysis of physical parameters such as soil properties, cli-mate, and hydrology, and the other based on discrimination by reference lines. Thefirst approach, physical characterization, requires access to substantial data sets and acapacity to interpret their implications for crop adaptation. It is hampered by a lack ofdata and poor understanding of plant responses to complex combinations of environ-mental factors. Since this is a priority topic for research, however, such relationshipswill become clearer as data coverage, methodologies, and physiological understand-ing improve (V.P. Singh et al, this volume).

In the second approach, differential responses of reference lines are used as abioassay for the occurrence of particular conditions (Cooper and Fox 1996, Wade etal 1996). Although this approach requires less environmental data for new test loca-tions, its effectiveness depends on knowledge of the reference lines and their patternsof adaptation. Confidence in the classification is improved and further understandingis developed if key environmental data are collected on the new test environments aswell as on cultivar response.

The availability of pattern analysis methods permits rigorous analysis of G × Einteractions and classification of genotypes and environments into groupings withsimilar patterns of adaptation or discrimination. Using these methodologies, recentstudies have reported a reliable assessment of repeatable G × E interactions in rainfedlowland rice (Cooper et al 1999, Wade et al 1999). Groups of varieties with commonpatterns of adaptation over environment groups were identified in these studies. Ref-erence lines would be chosen as standard and well-known representatives of thesevariety groups (Cooper and Fox 1996, Wade et al 1996). The principle is that a fewreference lines can be used to represent a wide range of genotypic adaptation. Fur-thermore, discrimination between these lines could be used to classify new test envi-ronments into previously identified targets.

An improved capacity to classify new test environments is important becausebreeders need to know how new sites relate to the target population of environmentsfor which they are breeding. Also, performance of a new genotype in a new set ofmultienvironment trials needs to be assessed relative to the performance of otherlines in similar environments. Both these requirements can be met by using a repre-sentative set of reference lines since the adaptation of a new genotype may be classi-fied by its similarity in response to the reference line with which it groups and byknowledge of the types of environments in which it and the reference line show simi-


lar adaptive responses. Thus, the availability of a known reference set of genotypescould greatly assist consideration of whether a new test environment is representativeof the selection target and whether or not a new genotype is preferentially adapted tothat target.

In addition to these immediate advantages for the interpretation of breeders’evaluations, we propose that reference lines be used to achieve a rapid and integrativeassessment of environmental characteristics over space and time and to indicate likelycharacteristics of new test environments. This would require evaluation of the refer-ence lines in many locations and seasons to sample a broad cross-section of possibleenvironments. For example, locations distributed across the toposequence would beneeded in order to assess the pattern of environments represented there.

Clearly, integration of physical and biological characterization is the ultimategoal for characterization, as this would lead to an improved assessment of whereparticular types of environments are present. The frequency and production risk asso-ciated with these environments could be assessed using crop simulation (Aggarwal etal 1996, Cooper et al 1999). Since physiological understanding of patterns of geno-type adaptation is required for the identification of useful traits conferring an adap-tive advantage in particular conditions, widespread use of a set of representative ref-erence lines could provide a first step in this process.

This chapter considers the feasibility and methodology for using a set of refer-ence lines to classify sites from multienvironment trials into an identified target popu-lation of environments. We consider strategies for selecting reference lines to capturea wide range of G × E interactions for this purpose. We examine methodologies forassessing the characteristics of new sites from multienvironment trials on the basis oftheir reference line responses. We seek to use these responses to indicate physical andclimatic properties of those new environments.


Using data from a series of 36 multienvironment trials with 47 entries conductedbetween 1994 and 1998 across South and Southeast Asia by the Rainfed LowlandRice Research Consortium and at the International Rice Research Institute, Wade etal (1999) reported that G × E interactions were repeatable and genotypes and environ-ments with similar patterns could be grouped consistently. The genotypes and envi-ronments they studied and their groupings and some class properties are listed inFigures 1 and 2. We seek to identify representative individuals from these genotypegroups for use as reference lines and to evaluate whether the G × E classification canbe repeated with just the reference lines. This would indicate whether a new environ-ment could be correctly classified based on reference lines alone.

In the selection of reference lines, we are concerned with sensitivity of responseof lines to environmental characteristics and would not like this sensitivity to be ob-scured by mean genotypic effects (McLaren 1996). Hence, we have used pattern analy-sis based on mean polished yield response data (genotypes and environments cen-tered with least squares adjustment for imbalance).


Fig. 1. Dendrogram of environment groups from pattern analysis of mean polished yield data. IND= India, THA = Thailand, PHL = Philippines, BGD = Bangladesh, IDN = Indonesia. (Reprinted fromWade et al (1999), Field Crops Research 64:35-50, with permission from the publisher.)

GrainSite Year Code Group yield Sand Clay Silt pH

(19xx) (t ha–1) (%) (%) (%)

Faisabad, IND 97 FC 3.10 35 16 49 7Udorn, THA 95 TD 63 1.90 50 20 30 6Phimai, THA 95 TE 0.88 43 40 17 5Phimai, THA 97 TK 3.11 46 38 16 5

2.25Raipur, IND 94 IA 46 1.26 15 51 35 7Raipur, IND 97 II 3.44 12 36 52 7

2.35Rajshahi, BGD 94 BA 2.26 28 44 28 6Rajshahi, BGD 96 BC 3.11 28 40 32 6Faisabad, IND 95 FA 2.77 35 16 49 7Faisabad, IND 96 FB 2.75 35 16 49 7Jagdalpur, IND 96 IG 62 2.43 31 44 26 6Raipur, IND 97 IH 0.11 15 50 35 7Jagdalpur, IND 97 IK 2.63 31 44 26 6Ubon, THA 94 TA 1.60 74 6 20 4Ubon, THA 95 TB 2.26 74 6 20 4Chumphae, THA 95 TC 0.21 60 32 8 4Chumphae, THA 96 TG 1.48 45 32 22 6Sakon Nakhon, THA 96 TH 1.50 76 7 17 4

1.93Rajshahi, BGD 95 BB 2.03 24 36 40 6Rajshahi, BGD 97 BD 2.06 28 40 32 6Raipur, IND 95 ID 0.38 18 51 32 8Raipur, IND 96 IE 57 0.80 16 53 31 8Jakenan, IDN 96 OB 1.78 26 12 62 6Ubon, THA 96 TF 1.24 74 6 20 4Ubon, THA 97 TI 2.29 74 6 20 4Chumphae, THA 97 TJ 1.78 45 32 22 6

1.54Jakenan, IDN 96 OA 60 4.62 26 12 62 6Jakenan, IDN 97 OC 3.50 26 12 62 6

4.06

Tarlac, PHL 94 PA 21 4.00 20 39 42 6

Raipur, IND 95 IC 10 2.55 15 53 33 7

Bilaspur, IND 97 IJ 16 2.51 15 53 32 7

Raipur, IND 95 IB 1.74 30 35 35 7Guimba, PHL 95 PB 1.89 34 20 46 7Tarlac, PHL 95 PC 61 2.11 20 39 42 6Muñoz, PHL 95 PD 2.38 33 47 33 7Tarlac, PHL 96 PE 0.92 37 21 42

1.81


Fig. 2. Dendrogram of genotype groups from pattern analysis of mean polished yield data. (Re-printed from Wade et al (1999), Field Crops Research 64:35-50, with permission from the pub-lisher.)

Grain Plant Days to Filled 1,000–Designation Code Group yield height flowering grain grain

(t ha–1) (cm) (%) weight (g)

Sabita 03 2.08 127 111 70.3 28.9Sabita-A 3A 2.15 128 114 65.4 28.3Sabita-B 3B 2.12 128 113 65.0 28.8KDML105 04 82 1.76 122 111 62.7 24.7KDML105-A 4A 1.76 125 111 60.2 23.9KDML105-B 4B 1.77 124 112 60.1 23.5IR57546-PMI-1-B-2-2> 21 1.50 122 116 61.2 23.3

1.88 125 112 63.5 25.9NSG19 02 2.12 126 96 69.8 27.1NSG19-A 2A 52 2.25 128 96 66.7 27.0NSG19-B 2B 2.21 126 96 67.8 27.1

2.19 127 96 68.1 27.1IR58821-23-1-3-3 22 1.56 110 121 54.5 23.1IR66469-17-5-B 25 78 1.76 113 108 53.6 21.7IR66516-11-3-B 27 1.64 110 113 51.3 24.4IR66516-24-3-B 28 1.91 116 114 54.6 24.5IR66516-37-7-B 29 2.00 113 109 64.1 23.0

1.78 112 113 55.6 23.4Mahsuri 05 2.00 106 109 70.2 17.4Mahsuri-A 5A 2.19 109 110 69.1 16.8Mahsuri-B 5B 79 2.15 110 110 71.2 16.9IR66883-18-2-B 36 1.76 102 111 62.3 22.9IR66883-18-3-B 37 1.75 103 111 61.2 24.0IR66883-44-3-B 38 1.86 108 111 57.4 21.9

1.95 106 111 65.2 20.0IR52561-UBN-1-1-2 15 1.90 118 97 57.9 21.0IR54071-UBN-1-1-3-1> 16 2.26 106 101 59.1 24.6IR57515-PMI-8-1-1-S> 20 2.67 107 103 66.8 27.0IR66506-5-1-B 26 2.01 99 107 58.5 24.3IR66879-19-1-B 30 84 2.07 105 105 56.1 20.5IR66879-2-2-B 31 2.11 103 106 58.4 21.1IR66879-20-2-B 32 2.18 103 108 57.0 20.5IR66879-8-1-B 33 2.01 101 108 56.0 20.5IR66883-11-1-B 35 2.15 111 99 66.0 26.0

2.15 106 104 59.5 22.8IR20 01 2.03 85 102 64.7 19.8IR20-A 1A 81 2.04 86 104 63.9 19.5IR20-B 1B 2.02 86 104 63.9 18.9CT9993-5-10-1-M 11 1.34 91 95 59.0 22.0IR66893-5-2-B 39 2.05 104 107 62.7 25.6IR58307-210-1-2-3-3> 62 2.24 91 109 63.7 18.5

1.95 90 103 63.0 20.7IR54977-UBN-6-1-3-3> 17 2.26 94 99 62.8 28.1IR57514-PMI-5-B-1-2 19 2.33 101 107 70.0 24.8IR62266-42-6-1 23 76 2.26 92 98 64.5 22.4IR63429-23-1-3-3 24 2.37 110 96 69.9 24.9IR66882-4-4-B 34 2.29 102 100 63.6 21.7

2.30 100 100 66.2 24.4PSBRc 14 08 74 2.71 83 89 69.9 21.3IR64 09 2.58 85 89 73.8 24.1IR36 10 2.61 74 91 71.6 22.4

2.63 81 90 71.7 22.6CT9897-55-2-M-3-M 12 2.46 86 97 68.4 23.0IR64615H 40 85 2.39 92 98 59.3 23.5IR68877H 42 2.63 87 89 62.9 22.7

2.49 88 95 63.5 23.1


Criteria considered for selecting reference lines were a high correlation withthe group pattern, a high yield level within the genotype group, a good knowledge ofcharacteristics of the line, seed availability, and, for the purpose of this simulationstudy, an adequate number of locations where the lines were tested.

The ability of a reference set to classify environments in the same way as a fullset of lines is assessed by comparing ordination (AMMI analysis) and cluster group-ings of the standardized reference line responses with those in the full analysis ofWade et al (1999). Two reference sets, one of eight lines and the other a subset of fivelines, are considered.

Three algorithms for classifying new environments on the basis of referenceline responses are compared: the expectation-maximization (EM) procedure imple-mented in the MATMODEL AMMI analysis package (Gauch 1990); the AMMI pro-cedure in the IRRISTAT G × E analysis module, which assumes zero interaction formissing values (IRRI 1998); and the pattern analysis procedure in GEBEI, whichuses a nearest cluster procedure to estimate missing values (De Lacy et al 1996). Tocompare the efficiency of the three algorithms, we used data for eight or five refer-ence lines over 36 environments. Yields in each of the 36 environments in the fullanalysis were omitted, one at a time, with just the reference line yields added back.The change in ordination position for the omitted location was used to assess theability of the method to accurately characterize a new environment on the basis ofreference line responses alone. The resulting classification may be used to indicatelikely characteristics of a new site and the types of genotypes that should be preferen-tially adapted there.


Selecting reference setsBased on their pattern analysis of the full data set, Wade et al (1999) suggested areference set containing representatives of six of the nine genotype groups detected intheir analysis. Their choice was based primarily on knowledge of the characteristicsof lines from the major groupings: Sabita or KDML105, NSG19, Mahsuri, IR57514-PMI-5-B-1-2 or IR62266-42-6-1, PSBRc14, and CT9897-55-2-M-3-M. Here, weconsider the criteria outlined in the materials and methods section as a basis for iden-tifying entries from the genotype groups to serve as reference lines.

In group 82 (Fig. 2), Sabita and KDML105 had similar patterns, but Sabita hada higher yield, so, from a practical point of view, it may be a better reference line. Forthis study, however, we chose KDML105 because it is widely grown in northeastThailand, a critical set of environments where drought and low soil fertility are com-mon.

There was no representative of group 78 in the original set of six indicated byWade et al (1999). The pattern for group 78 differed from that for 79 only at thesingleton environment group IC-Raipur 1995 (Fig. 3A, B). For this study, line 29(IR66516-37-7-B) was included to represent environment group 78.


Fig. 3. Interaction response patterns for four genotype groups. Genotype and genotype groupcodes are identified in Figure 2.

Group 84 was also not represented in the original set, so line 31 (IR66879-2-2-B) was selected on the basis of few missing values and good correlation with thegroup pattern (Fig. 3C).

No representative of group 81 was chosen because this group showed littleinteraction, low yield, and a weak pattern of discrimination between environments(Fig. 3D).

In group 76, line 34 (IR66882-4-4-B) was chosen as the reference line for thisstudy in preference to line 23 (IR62266-42-6-1) because the latter was evaluated atonly 27 of 36 sites. Line 34 had a reasonable correlation with the group pattern (Table1) and a high yield in its group (Fig. 2).

There was very little difference between PSBRc14 and IR36 in terms of patternor yield, with PSBRc14 preferred because it is recommended for rainfed lowlands inthe Philippines. Since PSBRc14 was not tested at all sites, however, IR36 was used inthis study to represent environment group 74.

The eight reference lines, whose characteristics are shown in Table 1, wereused in subsequent analyses in this chapter. To assess the effect of size of referenceset on reliability of characterization, a reduced set was defined by retaining only onerepresentative (IR66879-2-2-B) from groups 78, 79, and 84—the first to fuse in thestructure shown in Figure 2, and one representative (IR36) from groups 74 and 85—

Environment group

–3.0

2.5

1.4

0.3

–0.8

–1.9

GGP78

A

63 46 62 57 60 61 IC IJ PA

GTP28GTP29GTP27GTP25GTP22ZeroGGP78

2.5

1.4

0.3

–0.8

–1.9

–3.0

GGP84

C

63 46 62 57 60 61 IC IJ PA

GTP35GTP33GTP32GTP31GTP30GTP26GTP20GTP16GTP15ZeroGGP84

2.5

1.4

0.3

–0.8

–1.9

D

63 46 62 57 60 61 IC IJ PA

GTP62GTP39GTP11GTP1BGTP1AGTP01ZeroGGP81

–3.0

GGP81

–1.9

–3.0

2.5

1.4

0.3

–0.8

GGP79

B

63 46 62 57 60 61 IC IJ PA

GTP38GTP37GTP36GTP5BGTP5AGTP05ZeroGGP79


the next to fuse in the structure. This leads to a reduced group of five reference lines:KDML105, NSG19, IR66879-2-2-B, IR66882-4-4-B, and IR36.

Reliability of reference setsTo examine whether reference sets can reliably detect and estimate the same interac-tion as the full analysis, we repeated the pattern analysis reported in Wade et al (1999)using data for eight reference lines in 36 environments, then for only five referencelines in 36 environments.

Figure 4 shows the shift in environment ordination as a result of using data onreference lines alone compared with the full data set. With the full reference set (Fig.4A), movement is quite limited except for a few environments: IC, IJ, and TD. SitesIC and IJ nevertheless remain in similar positions relative to the remainder of theenvironments, so discrimination of these extreme sites (Wade et al 1999) is still rep-resentative of the full data set. Site TD is more of a concern, since it failed to distin-guish between the reference lines at all in this analysis, but discriminated strongly inthe full data set. This is due to specific interactions with a few test lines, which cannotbe captured by the smaller group. The average absolute shift due to using the refer-ence set of eight in place of the full set of 47 genotypes was 9% of the range in IPCA1scores and 10% of the range in IPCA2 scores. The use of the reduced reference set offive lines had more severe consequences, as can be seen by the relatively larger shiftsin Figure 4B, 13% for IPCA1 and 14% for IPCA2.

The clustering procedure on the full reference set of eight lines produced eightenvironment clusters that could be aligned with the original clusters in Figure 1. Table2 shows this. Of the 36 environments, 23 clustered into the same groups and 5 intoneighboring groups, leaving 8 or 22% that were not well characterized. With thereduced set of reference lines, 30% were not well characterized according to the fullanalysis.

Hence, reference lines are able to reproduce the main features of the environ-mental classification, but representatives of all the major genotype groupings were

Table 1. Characteristics of the reference lines chosen to representthe genotype groups shown in Figure 2.

CorrelationGenotype No. in Reference with group Range of

group group line pattern correlation

82 7 04 KDML105 0.81 0.59–0.9652 3 02 NSG19 0.92 0.92–0.9878 5 29 IR66516-37-7-B 0.92 0.92–0.9779 6 05 Mahsuri 0.95 0.56–0.9584 9 31 IR66879-2-2-B 0.87 0.51–0.9381 6 None –0.38–0.8576 5 34 IR66882-4-4-B 0.83 0.70–0.9274 3 10 IR36 0.98 0.93–0.9885 3 12 CT9897-55-2-M-3-M 0.93 0.83–0.93


Fig. 4. Environment ordination shifts due to using reference data. Letter codesfrom Figure 1 indicate environment positions from the full analysis, symbols at theend of the spokes in (A) indicate positions based on analysis of data from eightreference lines and in (B) for analysis based on five reference lines.

Table 2. Comparison of environment clusters formed by analysis of data from 8 reference linesover 36 environments and from the full data set from 37 genotypes over 36 environments.

Cluster tree from analysis Cluster number from full analysis (Fig. 2)

of reference lines

70 60 FC, TE,

TK

62 IA, BC BD

II

63 TD BA, FA, PA

FB, IH,

TH, TA,

71 69 66 TC, TG

64 IK, TTS BB, RE, OA IB

OB,TI,

65 TJ

IG ID IJ

61

68 IC

19

67 TF OC PE

59

PB, PC,

51 PD

63 46 62 57 60 21 10 16 61

B

FC

TETK

TD

IA

THTG

BB

IIFA

TFIDOC

PC

PDPE

OAPA PB

TO

IJIC

–1 0 1 2

TIBB

TA

IE

IPCA1

–2

2.0

1.3

0.6

–0.1

–0.8

–1.5

–2.2

A

–2 –1 0 1 2

TD

TKTE

FC

THTG

TBTFID

PC

OC

PDOA

PBPAIA

IK

TI

PE

BB

IC IJ

TA

IPCA2

FAII

IGOBBDTC IB


required as the quality of the reproduction dropped markedly when only a few groupswere represented.

Characterizing new environmentsIn order to check how “new” environments would be characterized on the basis ofreference line responses, environments were removed from the data set one at a time.For each removed environment, only responses of the eight reference lines were addedback to represent a “new” test environment. Pattern analysis on mean polished datawas carried out on each reconstructed data set and the shift in environment ordinationfor the new environment was recorded.

This technique will underestimate the error in characterizing truly new envi-ronments because the so-called “new” information had been included in the originalanalysis. It more accurately shows the effect of not having the responses on the testlines. The two concepts should not be too different, however, as the repeated informa-tion in each analysis is only eight responses in a matrix of l,692 cells.

Three analysis methods were used and two failed to characterize the new envi-ronments at all because their analysis relied heavily on the ability of a technique toimpute interaction effects for missing data. The more successful technique was toestimate the missing response of line i in environment j as the mean response inenvironment j of lines in the genotype group with which line i first fuses in the clusteranalysis. The resulting complete matrix is then subject to ordination (De Lacy 1996).There is no suggestion that this strategy is optimal, but it is certainly better than thestrategy of assuming zero interaction for missing lines as in the G × E module ofIRRISTAT (IRRI 1998), or the proportedly optimal EM strategy used in MATMODEL(Gauch and Zobel 1990), both of which lead to a complete collapse in the environ-mental characterization, with each replaced site shifting near to the zero interactionposition.

Figure 5 shows the results of the successful strategy. Shifts in the environmentordination are very modest. Mean absolute shifts were 8% of IPCA1 range and 11%

Fig. 5. Ordination shifts when each environmentin turn is represented only by data from refer-ence lines. Letter codes from Figure 1 indicateenvironment positions from the full analysis, sym-bols at the end of the spokes are the positionsbased on data from eight reference lines onlyfor that environment.

2.0

1.3

0.6

–0.1

–0.8

–1.5

–2.2–2 –1 0 1 2

IPCA1

IPCA2

IATD

TKTEFC

IITH

TGBA

TBFB

IG

FA PA PBPD

PCPE

IBOA

BB

OCTI

TI

IKTF

ID

BD

IC IJ

OBIE

BCIH

TA

TC


of IPCA2 range. The extreme, singleton environments IC and IJ showed large shifts,as did OA and TD, indicating that the characterization of these sites depended criti-cally on lines omitted from the reference sets.

Conclusions

The results indicate that a small set of reference genotypes is able to capture a signifi-cant amount of the G × E information that is available from a much larger set of testlines. The efficiency of the reference set depends, however, on having good represen-tation of all of the sensitive genotype groups.

The statistical methodology for this analysis needs to be improved as it relieson imputation of responses from one environment to another and there is no optimalway to do this at present.

Reference lines should be selected first on the basis of how well they match thediscriminatory pattern of the genotype group and then by desirable agronomic fea-tures, state of knowledge about the physiological responses of the line, and finallypractical considerations such as seed availability.

The reason for wanting well-known lines is that the ultimate goal of character-ization is to understand plant adaptation to different subecosystems in the target popu-lation of environments and to use this information to identify important traits thatconfer preferential adaptation to those target environments. From this knowledge,efficient selection strategies may be devised, but their success requires a detailedunderstanding of physical and climatic properties at all sites where reference sets aregrown. Similarly, detailed information on crop management and crop developmentare essential to make the ultimate link between adaptation and environmental condi-tions.

We propose that small trials of reference sets, widely grown in the rainfed low-lands, have the potential to provide a rapid and useful characterization of new envi-ronments. This can be used immediately to weight selection to the desired target en-vironments and to extrapolate the performance of test lines over wider geographicalareas. Integration of this classification with physical and climatic data would providethe link with a geographical characterization. The results can also be used to developan understanding of the physiological basis of plant adaptation, and, through a simu-lation analysis using historical climatic data, to complete the characterization in termsof frequencies of occurrence and risks associated with different target environments.

ReferencesAggarwal PK, Kropff MJ, Mathews RB, McLaren CG. 1996. Simulation models to design new

plant types and to analyze G × E interactions in rice. In: Cooper M, Hammer G, editors.Plant adaptation and crop improvement. Oxford (UK): CABI. p 403-418.

Cooper M, Fox PN. 1996. Environmental characterization based on probe and reference geno-types. In: Cooper M, Hammer G, editors. Plant adaptation and crop improvement. Ox-ford (UK): CABI. p 529-547.


Cooper M, Rajatasereekul S, Immark S, Fukai S, Basnayake J. 1999. Rainfed lowland ricebreeding strategies for Northeast Thailand. 1. Genotypic variation and genotype × envi-ronment interactions for grain yield. Field Crops Res. 64:131-151.

Cooper M. 1999. Concepts and strategies for plant adaptation research in rainfed lowland rice.Field Crops Res. 64:13-34.

De Lacy IH, Basford KE, Cooper M, Bull JK, McLaren CG. 1996. Analysis of multi-environ-ment trials: an historical perspective. In: Cooper M, Hammer G, editors. Plant adapta-tion and crop improvement. Oxford (UK): CABI. p 39-124.

Gauch HG Jr. 1990. MATMODEL Version 2.0: AMMI and related analyses for two-way datamatrices. Ithaca, N.Y. (USA): Cornell University.

Gauch HG, Zobel RW. 1990. Imputing missing yield trial data. Theor. Appl. Genet. 79:753-761.

IRRI (International Rice Research Institute). 1998. IRRISTAT for Windows tutorial manual.Biometrics Unit, The International Rice Research Institute, Los Baños, Laguna, Philip-pines.

McLaren CG. 1996. Methods of data standardization used in pattern analysis and AMMI mod-els for the analysis of international multi-environment variety trials. In: Cooper M, Ham-mer G, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 225-242.

Wade LJ, Sarkarung S, McLaren CG, Guhey A, Quader B, Boonwite C, Amarante ST, SarawgiAK, Haque A, Harnpichitvitaya D, Pamplona A, Bhamri MC. 1995. Genotype by envi-ronment interaction and selection methods for identifying improved rainfed lowlandrice genotypes. Proceedings of the International Rice Research Conference, 13-17 Feb-ruary 1995, International Rice Research Institute, Los Baños, Laguna, Philippines.p 885-900.

Wade LJ, McLaren CG, Samson BK, Regmi KR, Sarkarung S. 1996. The importance of envi-ronment characterization for understanding G × E interactions. In: Cooper M, HammerG, editors. Plant adaptation and crop improvement. Oxford (UK): CABI. p 549-562.

Wade LJ, McLaren CG, Quintana L, Harnpichitvitaya D, Rajatasereekul S, Sarawgi AK, KumarA, Ahmed HU, Sarwoto, Singh AK, Rodriguez R, Siopongco J, Sarkarung S. 1999.Genotype by environment interaction across diverse rainfed lowland rice environments.Field Crops Res. 64:35-50.

NotesAuthors’ address: International Rice Research Institute, DAPO Box 7777, Metro Manila, Philip–

pines.Acknowledgments: This chapter used data published by Wade et al (1999), which indicated the

contributions from scientists in the Rainfed Lowland Rice Research Consortium. Theseexperiments received support from the Asian Development Bank (ADB), DirectorateGeneral for International Cooperation (DGIS), and Department for International Devel-opment (DFID).


Biophysical characterization of rainfed systems . . . 143

Biophysical characterizationand mapping

144 Amien and Las


Limited suitable land for rice production in the past has forced the develop-ment of terraced and bunded rice fields at a higher elevation. Because therice-cropping system lacks terrestrial water resources, it totally depends onavailable rainfall. Java, North Sumatra, and South Sulawesi have the largestareas of rainfed lowland rice fields, 796,900, 210,300, and 259,100 ha,respectively. Because of high competition for space, rainfed rice area is steadilydeclining on Java but is relatively stable outside Java. A high proportion ofthe rainfed rice area is in the northern part of Java and eastern part of SouthSulawesi. The soils of rainfed areas in West Java were formed from acid tuft.In the other provinces of Java and South Sulawesi, the soils were formedfrom limestone or marine sediments. Less fertile soil that is low in P and Kand the low adoption of modern technology in rainfed areas resulted in lowerrice yields. Yields in rainfed areas were 10% to 25% less than the averageyield in Java and 15% to 20% less than the average yield in South Sulawesi.Yield levels of lowland rice on Java are inversely related to the proportion ofrainfed lowland rice areas. With monsoonal rainfall patterns, the rainy sea-son begins in October in West and Central Java, in November in East Java,and in March in South Sulawesi. West Java and South Sulawesi have nodistinct dry period, while in Central and East Java the dry period varies from4 to 5 mo. During El Niño years, the rainy season comes about a month later,with rainfall less than 70% of that of normal years. Dependence on rainfallalso makes rainfed rice more susceptible to drought, particularly during ElNiño years. This rainfed rice system helps reduce water runoff during therainy season, but differs with deep-rooted vegetation, because the waterretained is mostly lost through evapotranspiration. In the increasingly global-ized economy, labor shortages and more competitive markets threaten thesustainability of the rainfed rice system.

Biophysical characterization of rainfedsystems in Java and South Sulawesiand implications for researchI. Amien and I. Las

146 Amien and Las

As Indonesia’s staple food, rice is strategically important to the agricultural develop-ment and economy of the country. The crop is cultivated in diverse environments.Irrigated rice occupies the largest area, followed by rainfed lowlands, uplands, andtidal swamps, with 58%, 20%, 11%, and 11%, respectively. Because of less favorableenvironments, however, the yield and production of nonirrigated rice are far lowerthan those of irrigated rice.

The increasing demand for rice, along with the increasing population and lackof technological development in rice culture before the Green Revolution, has led toexpanded rice cultivation. With limited suitable land for ideal rice culture, new ricefields turned to less favorable environments.

Rice fields were developed on sloping land at a higher elevation, often withoutavailable water resources except rainfall. Terraced and bunded rainfed lowland ricehelps reduce water runoff and prevents flooding in the rainy season. But the systemcannot conserve water for the dry season because most is lost through evapotranspi-ration. When expanded further to even higher elevation and steeper slopes, the rainfedrice system disrupts the hydrological cycle in the watershed and becomes prone todrought in the dry season.

In the increasingly globalized economy with rapid progress in telecommunica-tions, transportation, and tourism, the low productivity of rainfed lowland rice, par-ticularly in Java, means less competitiveness. We therefore studied the performanceof rainfed lowland rice as affected by the physical characteristics of the environmentsin Java and South Sulawesi.

Rainfed lowland rice areas and their distribution in Indonesia

Figure 1 shows the distribution of rainfed rice area in Indonesia. Rainfed lowlandarea in Indonesia declined from 2.2 million ha in 1988 to 2.1 million ha in 1995 (BPS1988, 1995). There has been a slight increase in this area in Kalimantan and Sulawesi,but it is decreasing steadily in Java. The decreasing trend is also observed in Sumatraand Nusa Tenggara. The rate of conversion of rainfed lowland area to other uses ishighest in Java, with 13,800 ha annually, followed by Sumatra and Nusa Tenggara,with 6,100 and 2,700 ha per annum, respectively. The largest area of rainfed lowlandrice in the Outer Islands is in the province of South Sulawesi, with 259,100 ha or41.1% of the total rice area (Table 1). High proportions of rainfed lowland rice area inSouth Sulawesi are found in three districts in the east, Sinjai, Bone, and Wajo, with49.6%, 52.7%, and 79.6%, respectively, of the total rice area.

Although not quite as high in terms of the proportion to the total rice area,rainfed lowland rice area in the three big provinces of Java is among the largest in thecountry, with a total of 786,800 ha. Central Java has the largest area, with 293,600 haor 29.4% of the total rice area, followed by East and West Java, with 252,400 and240,800 ha or 22.0% and 20.9%, respectively, of the total rice area.

In Java, a large proportion of the rainfed lowland rice area is in the gently slop-ing northern coast. The districts of Lebak and Pandeglang in the northwest part ofWest Java have 43.1% and 45.9% of the total rice area. Grobogan, Blora, Pati, and


Fig. 1. Distribution of rainfed rice in Indonesia.

Table 1. Rainfed rice area in Indonesia.

Rainfed rice area (ha) ProportionYieldc

Region/province Single crop Double crop Drf/totala Rf/totalb (t ha–1)(%)

Sumatra 422,458 153,731 26.7 23.9 4.1Java 618,846 178,022 22.3 23.7 5.4

West Java 178,305 62,490 26.0 20.9 5.3Central Java 203,474 90,148 30.7 29.7 5.3East Java 229,783 22,623 9.0 22.0 5.5

Kalimantan 323,627 52,833 14.0 27.4 2.9Nusa Tenggara 47,872 6,391 11.8 13.8 4.5Sulawesi 258,785 39,118 13.1 31.6 4.4South Sulawesi 225,902 33,222 12.8 41.1 4.8

aRatio of rainfed double crop area to total rainfed area. bRatio of total rainfed area to total ricearea. cMean yield of all rice crops.

Rembang districts in the northeast of Central Java have 57.6%, 73.7%, 39.4%, and60.6% of the total rice area, respectively, in rainfed lowlands. All eight districts in thenorthern part of East Java, from Bojonegoro and Tuban in the west to Sumenep in theeast, have a high proportion of rainfed lowland rice area, which ranges from 41.3% inLamongan to 82.9% in Gersik.

Rainfed rice area (ha)

148 Amien and Las

Soil and climate of rainfed lowland rice areas in Java and South Sulawesi

SoilsThe soils of the rainfed areas in Lebak and Pandeglang districts of West Java wereformed from acid tuft. The soils developed from acid tufts in the region are classifiedas Inceptisols of the Dystropept subgroup or Ultisols of the Tropudults great group.They have low fertility although they have good physical properties (LPT 1967). Thesoils with a high clay content are able to retain the amount of water required byplants. This is particularly important when the impermeable plow pan in rainfed riceareas is not developed or is destroyed by soil tillage.

These soils are low in nutrients such as P and K as well as Ca and Mg. Underhigh iron and aluminum, the limited soil P is mostly retained by oxides. Althoughpuddled soil tends to release retained soil P, the low P potential in the soil means thathigh P fertilization is required to sustain high yield.

In the other provinces of Java and South Sulawesi, the soils were formed fromlimestone or marine sediments. Soils developed from these materials are classified asInceptisol of the Ustropepts great group, Alfisols of the Tropustalfs great group, andVertisols of the Chromustert great group (Dames 1955, LPT 1975). The smectiticclay minerals in the soils often make them difficult to till, particularly when water isinadequate early in the rainy season. The soils contain a relatively sufficient amountof Ca and Mg, but are low in P. The high bases often make K less available to plants.To attain higher yield, a higher rate of fertilization of P and K is required.

ClimateThe climate in the rainfed areas, because they are in the lowland tropics, is alwayswarm throughout the year. Differences in temperature are higher between night andday than between seasons. With ample solar radiation, the limiting factor for agricul-tural production is water availability. In these rainfed areas, the water supply totallydepends on rainfall.

Rainfall in Java and South Sulawesi is of the monsoon type, with distinct wetand dry seasons (Fig. 2). In a normal year, the rainy season with monthly rainfall of200 mm or more starts in September in the rainfed areas of West Java and in Novem-ber in northern East Java. During El Niño years from 1955 to 1994, the annual rainfalldropped about 30% with delayed rainy seasons. The rainy season in the eastern partof South Sulawesi starts in March during normal years, but is delayed until Mayduring El Niño.

Table 2 presents mean annual rainfall during normal and El Niño years in theregions with data from representative stations. From the number of wet months withmean rainfall of 200 mm or more only in Lebak and Pandeglang, West Java, the watersupply is adequate for more than two rice crops in one year, whereas in the other areaswet-season duration is only adequate for one rice crop. As indicated in Table 2, thereis a higher proportion of double crops in the rainfed areas of Sumatra and West andCentral Java, which have more wet months. During El Niño, water is available foronly one crop. In Madura (East Java) and South Sulawesi, rice growing is particularly


Fig. 2. Monthly rainfall distribution in rainfed lowland rice areas.

600

500

400

300

200

100

0

Rainfall (mm)

Rangkasbitung, West Java

Month

300

200

100

0

Gersik, East Java

J AJ F M A M J S O N D

400

300

200

100

0J F M A M J J A S O N D

Bone, South Sulawesi

400

300

200

100

0

Rembang, Central Java

EnsoNormalCrop 1Crop 2Crop 3

Table 2. Annual rainfall, wet and dry months, and month when rainy season begins in rainfedareas of Java and South Sulawesi.

Annual rainfall Wet Dry Month when rainyArea (station) months months season begins

Normal Ensoa

(mm) N E N E N E

West JavaRangkas 3,705 1,584 9 5 0 3 September November

Central JavaRembang 2,009 1,530 5 4 4 7 October December

East JavaGersik 1,621 1,103 4 4 5 8 November DecemberPamekasan 1,811 1,293 5 3 4 6 November December

South SulawesiWatampone 2,428 1,584 5 3 0 4 March May

aEnso = El Niño southern oscillation, N = normal years, E = years with El Niño.

150 Amien and Las

risky when the crop is planted late. Farmers are usually bound by tradition, and theyplant their rice crops as the rainy season begins in a normal year.

Effects of El Niño on rice harvest area

Because the water supply totally depends on rainfall, the climate has a very strongeffect on the rice harvest in rainfed lowland areas. Rainfed rice in West Java is usuallyplanted at the beginning of the rainy season from September to November and isharvested from January to April. The harvested rice area of West Java (Fig. 3) from1989 to 1996 indicates that rainfed areas in Lebak and Pandeglang districts sufferedmost during El Niño of 1991 and 1994. The harvested rice area of the first crop de-creased markedly from the normal year of 1990 to El Niño of 1991. In El Niño of1994, although the decrease in harvested area was not as high as in 1991, the droughtalso affected the harvested area in 1995. Although the harvested area of the first cropdecreased again in 1995, the total harvested area in 1995 increased significantly.

Fig. 3. Effect of climate on harvested rice areas in West and CentralJava.

Year (19..)

Central Java

1,600

1,400

1,200

1,000

800

60089 90 91 92 93 94 95 96

300

250

200

150

100

Harvested area (× 1,000 ha)

Grobogan, Blora, Pati,and Rembang

89 90 91 92 93 94 95 96

TotalFirst crop

160

140

120

100

80

60


Lebak and Pandeglang 3,000

2,000

1,000

0

West Java


In Central Java, the effects of El Niño of 1994 were more marked than those ofEl Niño of 1991. In Grobogan, Blora, Pati, and Rembang areas of Central Java (Fig.3), although the harvested area from 1990 to 1991 for the first crop showed a slightincrease, the total harvest that year decreased by almost 5,000 ha. But the decrease inharvested area in 1994 was almost 4 times higher for the first crop compared with thetotal harvest that year. In the rainfed lowland rice areas of Central Java, further de-creases in harvested area occurred after El Niño, as indicated by the reduction in areain 1992 and 1995 for the first crop harvest. The poor harvest of El Niño affected theharvest of the coming year probably because of the delayed planting time and un-availability of seed for planting.

In the eight districts of the north and in all areas of East Java, harvested areadecreased significantly during El Niño of 1991 and 1994 (Fig. 4). However, the de-crease was more pronounced in the eight districts of the north, which have a higherproportion of rainfed lowland area. The reduction in harvested area of the first cropfrom 1990 to 1991 in the eight districts accounted for 72.5% of the total reduction in30 districts of East Java. Although it was not as high as in 1991, there was a signifi-cant reduction in harvested area in El Niño of 1994, with a further decrease in 1995.

Fig. 4. Effect of climate on harvested rice area in East Java andSouth Sulawesi.

Year (19..)

300

200

100

0


Sinjai and Wajo

89 90 91 92 93 94 95 96

Central Java900

800

700

600

500

400

300

20089 90 91 92 93 94 95 96

TotalFirst crop

500

400

300

200


Eight districts ofnorthern East Java

1,600

1,400

1,200

1,000

800

600

East Java

South Sulawesi

152 Amien and Las

The decrease in harvested area only occurred in the first crop where rainfed lowlandrice was planted. The harvested area in the second and third crop increased slightly.

South Sulawesi has a different rainfall pattern (Fig. 2). Rainfed lowland rice iscommonly planted in March or April and harvested in August to September. Compar-ing the harvested area of the second crop harvested in August to December in Sinjaiand Wajo districts with the total harvest of the year showed that most of the cropfailure occurred in the second crop. This indicates that rainfed lowland areas sufferedmost from drought caused by El Niño. The decrease in harvested area of the secondcrop in the three eastern districts of South Sulawesi from 1990 to 1991 was more than70% of the total decrease that year (Fig. 4). A decrease in harvested area also tookplace in 1993.

The data from the four provinces showed significant adverse effects of climateon rainfed lowland rice production. Because of the total dependence on rainfall forwater supply, a reliable prediction of climate will help farmers plan better plantingtimes. A selection of early cultivars for El Niño years will also reduce the risk of cropfailure when the rainy season is shortened by one or more months.

Rice yield

The rice yield in districts with a high proportion of rainfed lowland rice area is alwayslower than the mean yield of the province or region. In Java, the rice yield of thepredominantly rainfed districts of Lebak, Rembang, Pati, Tuban, Bangkalan, Sampang,Pamekasan, and Sumenep from 1990 to 1996 was about 83% to 87% of the mean riceyield in Java (BPS 1990, 1991, 1992, 1993, 1994, 1995, 1996).

The lowest yield was in Lebak District, with a 7-year mean yield of 4.2 t ha–1

y–1 versus 5.5 t ha–1 in Java, probably because of poor soil conditions. Although ricetolerates acid soil and in puddled soils of paddy rice culture the pH of the soil movestoward neutrality, the high iron and aluminum in the acid soil retain soil P and the Pbecomes less available to plants.

In South Sulawesi, the rice yield from 1990 to 1995 in Sinjai and Wajo districtswas about 85% of that in all districts of South Sulawesi. With more adoption of mod-ern rice technology, the mean yield in the province steadily increased over time (Fig.5). But the mean yield in the districts with a high proportion of rainfed lowland rice,such as Sinjai and Wajo, fluctuated with interannual climate variability. The meanyield during El Niño 1994 was only 82.9% in Sinjai and 83.3% in Wajo of the meanyield in South Sulawesi.

When rice yield is plotted against the proportion of rainfed lowland rice area,there is a strong tendency indicating that the higher the proportion, the lower theyield. The 1996 rice yield of about 80 districts in Java plotted against the proportionof rainfed lowland rice area indicated a negative relationship with a linear equation of(Fig. 6)


Fig. 6. Relation between proportion of rainfedlowland rice area to total rice area and riceyield.

Fig. 5. Rice yield in South Sulawesi.

Yield (Qt ha–1) = 55.963 – 0.137* proportion (in % of total rice area)

Labor requirement and technology adoption

A study in Lampung, Java, and South Kalimantan by Kasryno and Sudaryanto (1994)reported that rice farming in rainfed lowlands required 175 man-days from land prepa-ration to harvest compared with only 157 man-days in irrigated areas and 71 man-days in tidal swamp areas. Compared with irrigated rice, rainfed rice requires morelabor for land preparation, seeding, transplanting, and crop care but less for harvest.Land preparation, seeding, and transplanting probably require more labor because ofthe difficult terrain and harder-to-till soil. Land preparation is particularly difficult inswelling-clay soil types such as Vertisols that require adequate water for land prepa-ration.

Being somewhat remote or less accessible because of the distance from eco-nomic centers, rainfed lowland rice areas took longer for the adoption of modern ricetechnology. In 1970, only 8% of the rainfed rice area used modern rice varieties com-pared with 53% of the irrigated rice area (Kasryno and Sudaryanto 1994). The pro-portion of modern rice varieties increased to 42% in 1980 and 81% in 1987, whereasin the irrigated rice system it was 91% in 1980 and 98% in 1987.

Relatively poor soil conditions like those in rainfed lowland rice areas willrequire more fertilizer to attain a higher yield. Fertilizer application in rainfed low-land rice in 1987 as reported by Kasryno and Sudaryanto (1994) was only 107 kgha–1 of a combination of urea, triple superphosphate, and KCl fertilizers, versus144 kg ha–1 in irrigated rice. Much more fertilizer was applied in rainfed lowland rice

45

43

41

39

37

35Sinjai Wajo South

Sulawesi

Yield (Qt ha–1)

District

Y95 Y92

Y94 Y91

Y93 Y90

Yield = 55.963 – 0.137* PR = 0.66**

Rainfed proportion (%)

65

60

55

50

45

400 10 20 30 40 50 60 70 80

Yield (Qt ha–1)

154 Amien and Las

than in tidal swamp rice, for which only 56 kg ha–1 was applied. The slower adoptionof modern rice technology coupled with less favorable environments in rainfed low-land rice areas have resulted in a lower rice yield.

Research priority and policy implications

Although the trend is declining, rainfed lowland areas in Indonesia, with more than 2million ha, play an important role in producing staple food and providing employ-ment opportunities for the population. These rainfed lowland areas are commonlysituated at a higher elevation in the watershed, but are also scattered in small patcheswherever human settlements exist. To better understand the nature and characteristicsof rainfed lowland rice, a systematic program to delineate its area is required. Thedelineation of rice area in Java using remote-sensing technology needs to be furtherexpanded to cover other areas in the country.

The lower yield of rice in rainfed lowlands occurs mainly because of the sloweradoption of modern rice technology. Less favorable soil conditions can be overcometo some extent by applying the appropriate type and rate of fertilizers. Research onfertilization for rainfed lowland rice and subsequent crops in an appropriate croppingpattern for the region needs to be pursued. In the more humid areas, such as WestJava, in a normal year, rainfall is adequate for two rice crops, which can be followedby a less water-demanding secondary crop. In the subhumid areas of Central and EastJava, rainfall is adequate at least for one rice crop.

The susceptibility of rainfed lowland rice to drought caused by interannual cli-mate variability has affected it more than other rice cultivation systems. With a betterprediction of climate, better planning for time of planting and selection of rice culti-vars can be achieved to avoid crop failure. Research to improve the accuracy of sea-sonal climate forecasting is currently being done by national and international agri-cultural research institutions. This cooperation needs to be further promoted for abetter understanding of the dynamics of tropical climate.

ReferencesBPS (Central Bureau of Statistics). 1988. Luas dan Penggunaan Lahan di Indonesia. Jakarta

(Indonesia): BPS.BPS (Central Bureau of Statistics). 1990. Agricultural survey: production of paddy in Indone-

sia. Jakarta (Indonesia): BPS.BPS (Central Bureau of Statistics). 1991. Agricultural survey: production of paddy in Indone-




sia. Jakarta (Indonesia): BPS.


BPS (Central Bureau of Statistics). 1995. Agricultural survey: production of paddy in Indone-sia. Jakarta (Indonesia): BPS.

BPS (Central Bureau of Statistics). 1996. Agricultural survey: production of paddy in Indone-sia. Jakarta (Indonesia): BPS.

Dames TWG. 1955. The soils of East Central Java. Contributions of the General AgriculturalResearch Station, Bogor, Indonesia.

LPT (Lembaga Penelitian Tanah). 1967. Reconnaisance soil map of Java. Bogor (Indonesia):LPT.

LPT (Lembaga Penelitian Tanah). 1975. Reconnaisance soil map of South Sulawesi. Bogor(Indonesia): LPT.

Kasryno F, Sudaryanto T. 1994. Modern rice variety adoption and factor-market adjustments inIndonesia. In: David CC, Otsuka K, editors. Modern rice technology and income distri-bution in Asia. Boulder, Colo. (USA): Lynne Reinner Publishers, Inc., and Manila (Phil-ippines): International Rice Research Institute. p 107-127.

NotesAuthors’ address: Center for Soil and Agroclimate Research, Bogor, Indonesia.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-


Monitoring rainfed and irrigated rice in Southeast Asia . . . 157

Monitoring rainfed and irrigated ricein Southeast Asia using radar remotesensingR. Verhoeven, H. van Leeuwen, and E. van Valkengoed

Rice is the main food crop in the Asia-Pacific region and reliable spatial infor-mation such as area, yield, crop type, and secondary crops is therefore es-sential for management of the rice-growing areas and food policy.

Space-borne remote sensing provides a synoptic view over extensiveareas on a regular basis and has proven to be useful as an independent datasource for agricultural statistics. Radar remote sensing can be useful formapping and monitoring cloud-covered areas and can be used to identify andmonitor rice fields. This is demonstrated on the basis of some current andpast projects on rice in Southeast Asia. Using an integrated spatial approach,combining information from remote sensing and other spatial data sourcesin a geographic information system, rice crops can be identified in both thedry and wet season and useful information such as area, number of har-vests, and eventually yield can be estimated.

In the European context, European Union-Directorate General I (EU-DG-I) and EU-DG-VI have begun two major rice programs to support food security monitoring inrelation to the rice-cropping systems in Asia using a weather-independent and remotesensing-based approach. In the philosophy of the European MARS program (“Moni-toring Agriculture with Remote Sensing”) to produce crop statistics on the nationallevel, the Southeast Asian Rice Radar Investigation (SEARRI) and Satellite Assess-ment of Rice in Indonesia (SARI) programs have been defined for rice mapping andproduction monitoring using radar remote sensing and national rice statistics fromthe International Rice Research Institute (IRRI) and the national government of Indo-nesia, respectively. The European Space Agency (ESA) has begun several data userprogram projects (DUP) using the latest developments in radar remote sensing inrelation to rice-wheat-water in Sri Lanka, Europe, and Bangladesh.

Synoptics Remote Sensing and GIS Applications BV is involved in some majorrice investigation, monitoring, and yield-forecasting projects. The current projects inSoutheast Asia where Synoptics plays the role of information facilitator in rice map-

158 Verhoeven et al

ping and monitoring using spatial techniques (remote sensing and geographic infor-mation system, GIS) are the following:

1. SEARRI project (1998-99): The Southeast Asian Rice Radar Investigationproject is a demonstration project within the Centre of Earth Observation(CEO) program AS3200 package (CEO, Aschbacher 1995, Division of theJRC-EU, Italy). The SEARRI technical objectives were consolidated in re-sponse to the requirements expressed by several Directorate Generals withinthe European Commission, in particular DG-VI, DG-IX, and DG-I-B. Theproject outcome is meant to support the DGs’ requirements. The main scopeof the project is to map at a regional scale rice cultivated areas in SoutheastAsia using space-borne synthetic aperture radar (SAR) and GIS technology.The SEARRI thematic products are intended to support two main objec-tives: crop production and methane emission estimations. The first theme isrelated to important socioeconomic issues, rice being the dominant sourceof food in Southeast Asia; the second is linked to climate and global changeissues, methane being one of the important greenhouse gases.

2. SARI project (1998-2001): The Satellite Assessment of Rice in Indonesiaproject is defined by the Indonesian government and European Union-DG-I-B and the Joint Research Centre (JRC) in Ispra, Italy, to improve the abilityto accurately and timely predict rice yields in Indonesia. Furthermore, theSARI project is defined to meet the wish of the Indonesian government todevelop an independent and efficient rice production-forecasting system simi-lar to what is being implemented in Europe for a range of crops in the MARSproject. The set-up of the project is to adapt MARS techniques to the Indo-nesian situation and conditions. Earth observation satellites using activemicrowave sensors (radar), such as the current European ERS and futureENVISAT, are unhampered by the presence of clouds. Information fromthese types of satellites would permit regular observation of the rice-grow-ing cycle and in principle could form the backbone of an independent moni-toring system adapted to the Indonesian context.

3. UPRICE project (1998-99): This project is funded by the Dutch NationalRemote Sensing Programme (NRSP-2) of the Netherlands Remote SensingBoard (BCRS) and aims to provide and upscale rice location/flooding statusinformation from satellite radar images of local fields to the regional scaleof Luzon Province in the Philippines. It has been demonstrated in variousstudies (Kurosu et al 1995, Le Toan et al 1997) that satellite synthetic aper-ture radar data from the ERS and JERS instruments can be successfully usedto monitor rice growth. Two sensors on board the ERS satellite have beenused in this project: the SAR sensor and the windscatterometer. The SARimages were used for detailed mapping of rice crops, whereas thewindscatterometer gave a coarse representation of the area at a high revisit-ing frequency. Once adequately validated with detailed SAR images,windscatterometer data may provide a handle on the dynamics of flooding


and, indirectly, methane emission from rice-wetlands at subcontinental scales(Denier van der Gon et al 1999, van Leeuwen 1998).

Classification of rice using radar image time series

SAR remote sensing offers favorable capabilities for rice mapping and monitoring.The typical radar response to rice fields at different growing stages is the key factorfor discriminating rice from other land-use classes. Le Toan et al (1997) studied theradar response to rice fields at different growing stages. They distinguished threemain periods in the crop cycle: (1) sowing-transplanting, (2) growing, and (3) theafter-harvest period. They found a bridge-shaped backscatter pattern representing theradar signature of rice during the growing period (Fig. 1). In ERS, the SAR timeseries classification of rice can take place on the basis of this typical signature. LeToan demonstrated that detection of changes of 3 dB or more over the growing seasonindicates rice areas.

The methodology of Le Toan works fine for extensive rice areas in flat terrainin the same growing stage. For smaller rice fields, however, this methodology doesnot work, especially not in mountainous terrain, where the dynamic range of back-scatter is somewhat compressed (Verhoeven and van Leeuwen 1999). Also, the pres-ence of small rice fields in different growing stages is difficult to detect using thismethod.

For classification of rice using radar time series, standard methodologies suchas maximum likelihood of iso-data clustering can yield unsatisfying results. A majordisadvantage of these techniques is that they ignore the typical time sequence of the

Fig. 1. Radar backscatter response over a rice-growing cycle. Source:CEO-EWSE.

Sowing

Tillering

Stem extension

Radar backscattering

Surfacecontribution Interaction

Volumecontribution

Vegetative stage Reproductive stage

Sow

ing

Tille

ring

Ste

mex

tens

ion

Flow

erin

g

160 Verhoeven et al

radar signature of rice as well as the spatial coherence of neighboring pixels. Classi-fication can be performed more efficiently if nonrice classes such as urban settle-ments, layover, water, and other crops are masked out beforehand. A methodologyapplied by Le Toan et al (1997) uses multitemporal changes of 3 dB or more as a ricemask. As this methodology is very sensitive to speckle, a multitemporal speckle re-duction filter should preferably be applied to preserve image geometry.

The ERS SAR sensor has a spatial resolution of 25 m and acquisitions areseparated by 35 days in time, so three to four consecutive ERS scenes cover a typicalrice crop cycle. Many rice paddies are small and of the same order of magnitude asthe sensor resolution; they are therefore hard to detect. Another effect called “speckle”is typical for SAR imaging and hampers classification on a pixel level. It is clear that,for a proper classification of rice, additional information is necessary. Knowledge oftransplanting/sowing date, type of cultivar, land use, soil type, and water supply (irri-gated, rainfed) is therefore essential. Here, a GIS could serve the purpose. Validationof the classification results can take place using existing agricultural statistics andfield survey data. Though accurate determination of rice on a field level is difficult oreven impossible to achieve with most space-borne remote-sensing images, reliablestatistics at larger geometric scales (e.g., on a district or provincial level) can be de-rived.

Low-resolution data with a high repetition frequency can cover gaps in time. Inspite of a lack of geometric detail, the overall rice signature in homogeneous areascan still be seen at a large scale. Important for classification of rice paddies is themoment of transplanting or sowing. Because of the time gap of 35 d for the ERS SARsensor, this moment has not always been monitored. Low-resolution data can be ofgreat help, especially here, to improve classification at lower scales. When studyingthe cropping cycle of rice in, for example, the Philippines (UPRICE project, Deniervan der Gon 1996, Denier van der Gon et al 1999), there is for each growing cycle aperiod of about 4 to 6 wk of flooding and after that the rice crop matures into a fullcoverage within several weeks. These dynamics can be followed by the radar and candistinguish the rice crop from other crops having different crop management (see Fig.2). Even low-resolution radar data (up to 50 km!) can result in these signatures on aregional scale when the rice crop (and related hydrological features) is dominantlypresent in the region.

In the projects mentioned, an iterative integrated process has been set up toimprove the classification of rice. The basis for a proper classification is a co-regis-tered and speckle-filtered set of ERS SAR images. An optimal image set can be se-lected on the basis of cropping calendars. Possible rice areas are selected on the basisof their dynamic behavior in time. Stable objects, such as forests, will be excluded inthis phase. Vector maps can also be used to mask out nonrice areas (e.g., open water,urban areas). Supervised or unsupervised classification of the remaining areas is nowmore efficient and the presence of probable mixed classes will be reduced. Afterclassification, analysis of the radar signature as a function of time is used to mergeclasses that have a similar signature and are geometrically linked to each other. Anexample of a radar backscatter signature of rice can be seen in Figure 3. The latter can


Fig. 2. The cropping calendar of rice in Luzon Province of the Philip-pines and the rice-specific radar signature in time of low-resolutionradar satellite data. The so-called “upscale moments” indicate analmost simultaneous acquisition of ERS SAR and windscatterometerdata.

Fig. 3. Radar backscatter signature for 2-cycle rice in the MekongDelta, Vietnam. Horizontal axis represents acquisition time (1996),vertical axis is radar backscatter in dB. The vertical bars representone standard deviation.

–7

–8

–9

–10

–11

Radar backscatter (dB)

Upscale moments:Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov DecDry-season rice crop Wet-season rice crop

ERS windscatterometer dynamics1995

0

–5

–10

–15

–20

dB2-cycle rice flooded in May and September

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Month

162 Verhoeven et al

be done using automatic or manual clustering techniques, such as grouping of radio-metric and geometrically related areas in the image series. Existing vector maps froma GIS can also be used for grouping similar areas.

Ancillary spatial information such as rice maps, rice statistical data, land-usemaps, crop calendars, and field data can be of great use for improving and refining theclassification. Knowledge about transplanting or sowing date is very important fordiscrimination of rainfed crops from other rice crops.

The SEARRI project demonstrates the integration of rice information derivedfrom radar observations, statistics, expert knowledge, and land-use information in aGIS environment. End-users could have access to the database yearly. The areas cov-ered by radar data combined with the occurrence of rice according to the Hukes’database (Huke and Huke 1997, IRRI 1995) can be seen in Figure 4. Figure 5 showsthe workflow followed in the SEARRI project, where time series of radar (ERS)images are combined with regional statistics, expert knowledge, crop calendars, andmeteorological data in a GIS environment. The radar images were all acquired be-tween March 1996 and December 1997. The thematic rice maps can be validatedagainst statistics (Hukes’ database) and land-cover maps in the tailor-made SEARRIGIS application. In total, 261 ERS scenes divided over 52 locations (3 to 9 images perlocation) were processed. The ERS time series covers almost all major rice-growingareas in Southeast Asia, including Thailand, Cambodia, Vietnam, and Malaysia.

In the UPRICE project, crop calendars and rice statistics on the municipalitylevel were available. The radar data consist of a time series of five ERS images of1995. The time series of ERS images was classified using a supervised classification

Fig. 4. Location of ERS scenes selected for SEARRI in combinationwith provincial rice crop statistics from the Hukes’ database.


methodology. Areas within the municipalities with near-100% rice cultivation wereused as training data.

The SARI project (still running at the time of the writing of this chapter) aims atassessing rice yields on a regular basis for all of Indonesia, similar to the EuropeanMARS program. Because most paddy fields are smaller than the agricultural areasmonitored in the MARS program, an integrated remote-sensing and GIS environ-ment similar to the SEARRI approach will be necessary. To validate biomass andyields, an intensive field campaign runs parallel to the satellite acquisitions.

ERS SAR time series, acquired every 35 days, are used to classify rice paddiesand the number of crops and to estimate area and biomass. Classification of rice areasis knowledge-driven, using GIS data based on the methodology of SEARRI. Cropgrowth models, such as ORYZA, use data from different spatial sources (e.g., fertil-izer information), biomass estimates, and crop cycle information from the remote-sensing data.

Low-resolution imagery such as NOAA/AVHRR (1-km resolution), SPOTVÉGÉTATION (1 km), RADARSAT ScanSAR (100 m), and, in the near future, theENVISAT ScanSAR (150 m) is being used as well. These tools can deliver usefulinformation on a more regular basis than ERS SAR does and can be used to controlthe high-resolution classification process as well. An example of the usefulness oflow-resolution remote-sensing data is the rice yield–forecasting system SHIERARY,developed by SEAMEO BIOTROP in Indonesia, which provides rice yields on a

Fig. 5. SEARRI image-processing workflow using remote sensingand GIS data to identify rice fields.

ERS SAR imageseries

Texture andspeckle-filtered

images

Cropping calendarsMeteorological data

Rice maps

Final classifiedgrid

2. RSpreprocessing

3. RSpostprocessing

4. GISanalysis/

presentation

Regionalstatistics/maps

Thematic maps(rice versus nonrice,single versus doublecrop, planting dates)

Hukes’statistics

Administrativegeodataset

ERS-frames

1. GIScombination/

selection

Geodatasets onHukes’ statistics

Seasonalland-coverregions

Expertknowledge

Validationcomments

164 Verhoeven et al

provincial level using daily data from the NOAA/AVHRR sensor. Integration of dif-ferent data sources will lead to a reliable rice-forecasting system.

For the SARI project, the acquisition of ERS images is accompanied by radar-dedicated field surveys, in which information about the field, crop(s), soil, hydrologi-cal and meteorological conditions, and other parameters is acquired to validate classi-fication results from the ERS image series and to calibrate biomass estimates fromthe radar data. Table 1 gives an example for field parameters assessed.

Stratified sampling of rice fields in Indonesia for suitable and less suitable riceareas is performed. This approach is not synchronized with the radar image acquisi-tions. Using aerial photos over 1 × 1-km areas and the actual situation in the field,plots with similar phenological stages or land-cover type are being digitized and storedin a GIS. Information from this stratified sampling survey is meant to give anotherindependent rice data source, and can be used to validate crop area from the remote-sensing data (Table 1).

Some results

The classification results of the UPRICE project can be seen in Figure 6. The classi-fication accuracy is 78% when comparing the results with the figures of the Provin-cial Bureau of Regional Statistics. The rice map indicates the rice-growing area dur-

Table 1. Example of form used to collect ground information for the SARI project.

LAND PREPARATION +WIND NO GROWTH STAGE R

RAIN NO HOMOGENEITY OF FIELD 3

PESTS/DISEASES NO PLANT ROW DIRECTION NW

INTERCROPS NO SURFACE SHAPE OF FIELD –9.00

INTERCROP NAME NA SURFACE SHAPE OF WATER –9.00

LAND USE PADI STRUCTURE OF PLANT 1+4

RICE VARIETY LOKAL SOIL MOISTURE 4

UPLAND NO MAXIMUM PLANT HEIGHT 105.20

IRRIGATED YES AV OF LEAVES PER PLANT 43.00

RAINFED NO AV OF STEMS PER PLANT 29.00

TIDAL/SWAMP NO AV LENGTH OF LEAVES 29.95

SEEDED NO AV WIDTH OF LEAVES 1.39

TRANSPLANTED YES AV NUMBER OF PANICLES 28.80

DATE OF TRANSPLANTING 9-4-99 AV PLANT DENSITY 16.00

YIELD –9.00 AV WATER LAYER –9.00

FERTILIZER YES AV BIOMASS OF STEMS AND 136.20

LEAVES

FERTILIZER NAME UREA, SP36 AV WEIGHT OF PANICLES 108.00


Fig. 6. Supervised classification of ERS time series, Luzon Province,Philippines. Classes are rice (yellow), urban area (red), layover (white),radar shadow (dark gray), rivers/open water (blue), and other (green).

ing the wet season. In a second phase of the project, ERS radar data of the dry seasonare included to detect the dry-season rice areas (irrigated rice).

Figure 7 (part of) shows the classification results of the SEARRI project com-bined with administrative boundaries. Validation is mainly done with the help of theIRRI provincial database on rice of the region (Hukes’ database). Although the Hukes’database estimates are only valid for the base year not later than 1995, it is the onlydatabase available at the required scale (covering Southeast Asia). In general, thetotal rice area from the radar classification is in accordance with the Hukes’ database(± 70% accuracy).

Improvement of the rice map is needed for the occurrence and number of ricecycles throughout the season. Because of the heterogeneous spatial character of manyrice-growing areas, it is still difficult to detect all rice cycles using a limited numberof radar images.

Using the area and number of cycles throughout the season from the classifica-tion results, rice production and methane emission maps are generated. Figure 8 showsthe yearly rice production per province.

166 Verhoeven et al

Fig. 7. A part of the SEARRI final radar classification in a GIS layer com-bined with administrative boundaries. The map is based on radar time se-ries acquired between March 1996 and December 1997. The red area inthe middle of the image is the city of Bangkok, Thailand.

Fig. 8. Rice production map for the major rice-growing areas in SoutheastAsia (t y–1).


Conclusions

Monitoring of rice crops in the Asia-Pacific region in terms of changes in productionand quality can be supported with a combination of radar remote sensing and GIS.Radar remote sensing provides an all-weather monitoring tool and, using time seriesof radar data, useful information about rice, such as area, growing stage, and cyclingof crops, can be assessed over extensive areas and rice can be identified even at largegeometric scales.

Despite the advantages of radar remote-sensing satellites, in the case of verysmall fields or mountain slopes, classification of rice will be less accurate. Integrationof different spatial data sources using a GIS in combination with crop growth modelswill improve the identification of rice and assessment of area, biomass, and yield.Low-resolution optical imagery (NOAA AVHRR, SPOT VÉGÉTATION) and radar(ScanSAR sensors) can improve the identification and classification of rice areas at alarger scale.

ReferencesAschbacher J. 1995. Rice mapping and crop growth monitoring, an ERS/SAR demonstration

project. Earth Observ. Q. 49:1-3.Denier van der Gon HAC. 1996. Methane emission from wetland rice fields. PhD thesis.

Wageningen Agricultural University, Wageningen, Netherlands. 182 p.Denier van der Gon HAC, Janssen L, van Leeuwen HJC, Verhoeven R, van der Wal JT, van der

Woerd H. 1999. Upscaling methane emissions from wetland rice fields (UPRICE). Po-sition paper on the feasibility of employing radar data for the Philippines. Interim reportBCRS Project 4.2/AP-09.

Huke RE, Huke EH. 1997. Rice area by type of culture: South, Southeast, and East Asia, arevised and updated data base. Los Baños (Philippines): International Rice ResearchInstitute. 59 p.

IRRI (International Rice Research Institute). 1995. Field variabilities of soil and plant: theirimpact on rice productivity and their use in modelling of soil kinetics and rice yield.Terminal report 1992-1995. The International Rice Research Institute (IRRI) & UniversitätLeipzig, Los Baños, Philippines.

Kurosu T, Fujita M, Chiba K. 1995. Monitoring of rice crop growth from space with ERS-1 C-band SAR. IEEE Trans. Geosci. Rem. Sens. 33(4):1092-1096.

Le Toan T, Ribbes F, Floury N, Wang LF, Ding KH, Kong JA, Fujita M, Kurosu T. 1997. Ricecrop mapping and monitoring using ERS-1 data based on experiment and modellingresults. IEEE Trans. Geosci. Rem. Sens. 35(1).

van Leeuwen HJC. 1998. Feasibility study in using scatterometer data for wetland rice map-ping yielding methane emission indicators for global applications. Proceedings of a JointESA-EUMETSAT Workshop on Emerging Scatterometer Applications from Researchto Operations, 5-7 Oct. 1998, at ESTEC, Noordwijk. ESA-SP-424.

Verhoeven R, van Leeuwen H. 1999. Satellite assessment of rice in Indonesia (SARI). Interimreport, April 1999.

168 Verhoeven et al

NotesAuthors’ address: SYNOPTICS Remote Sensing & GIS Applications BV, Costerweg

1-k, 6702 AA Wageningen, The Netherlands, Tel: +31 317 421221, Fax: +31 317 416146,E-mail: [email protected], [email protected],[email protected], [email protected].

Acknowledgments: The participating organizations in the Asia-Pacific region, the EuropeanUnion, European Space Agency (ESA), and Netherlands Remote Sensing Board (BCRS)are kindly acknowledged for subsidizing the abovementioned studies. The SARI projectand its staff, Ir. Mubekti in particular, are gratefully thanked for their valuable contribu-tions to this publication and its presentation at the IRRI workshop in Bali.


Characterizing soil phosphorus and potassium status . . . 169

Mapping of extractable soil phosphorus (P) and potassium (K) began in Indo-nesia to define soil fertility in lowlands and uplands that were under rice orlikely to be brought under cultivation. These maps, however, have not beenpublished and have therefore been largely inaccessible to researchers out-side of Indonesia. This chapter reviews the history of soil P and K map pro-duction in Indonesian lowlands and uplands, how mapping has been used toimprove P and K fertilizer recommendations in mapped regions, and limita-tions of the present mapping approach in determining accurate P and K fertil-izer requirements.

The latest edition of soil nutrient status maps produced in 1998 coversall the lowlands of Java, Lampung in southern Sumatra, and some of thesmaller outer islands with a scale of 1:250,000 or 1:500,000. The soil Pand K status of the lowlands has been defined as low, medium, or highbased on the responsiveness of rice crops grown in field trials to fertilizer Pand K. Trials and soil surveys have shown that there is a good relationshipbetween expected P and K requirements and actual P and K responses inlowland rice grown in soils with a low P and K fixation capacity. The maps,however, have limited application in lowland areas with high fixation capacityand their low resolution can lead to fertilizer recommendations being inaccu-rate. Despite these limitations, soil P and K mapping in lowlands has re-sulted in the blanket recommendations given in the 1960s being replaced byrecommendations specific for small regions within provinces.

Few extractable soil P and K maps have been developed for upland rice-cropping regions. The usefulness of soil P maps as a tool for developingfertilizer recommendations is limited in uplands because many soils havehigh P sorption capacity, particularly in Lampung. Fertilizer recommendationsfor upland rice cropping are still broad, especially for K.

Characterizing soil phosphorusand potassium status in lowlandand upland rice-cropping regionsof IndonesiaA. Clough, I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno, and S. Fukai

170 Clough et al

The Centre for Soil and Agroclimate Research (CSAR) has been conducting a long-term project on mapping of extractable soil P and K in lowlands since the late 1960s.Extensive mapping efforts have been restricted to soil P and K in lowlands becauselowland rice is the most economically important crop production system, and P and Kare the nutrients most likely to increase grain yields after nitrogen. The work at CSARhas resulted in complete extractable soil P and K maps for all the rice-cropping low-lands of Java, Lampung in southern Sumatra, Bali, and Lombok. Some small uplandareas have also been mapped by CSAR for extractable soil P and K as part of landassessment processes conducted for the transmigration programs of the 1980s.

Nutrient status maps produced solely by CSAR have not been published andhave therefore been largely inaccessible to researchers outside of Indonesia. Thischapter aims to introduce the evolution of extractable soil P and K mapping in Indo-nesia to the wider scientific community and demonstrate how the maps have beenused to improve P and K fertilizer recommendations. In conjunction with mappingexercises, many fertilizer response trials have been conducted, sometimes in the samelocation where soil samples were taken for mapping. Results from the field trialshave been used with the maps to create fertilizer recommendations for small regions.

The review primarily examines the maps produced solely by CSAR but alsoincludes higher resolution maps produced by other institutions to highlight the limita-tions of low-resolution mapping in improving fertilizer management. This review ofthe relationship between mapping and fertilizer recommendations is restricted to Javaand Lampung because these areas are the predominant ones for rice cropping andthey have high variability in P and K status relative to other mapped regions.

The relationship between soil P status in the maps and P responsiveness of riceis exemplified through field trials including some trials conducted in 1998 in Lampung.Lampung was selected because the province has been mapped and it produces ricecrops that are responsive to P but not always as expected from soil P maps because ofP fixation. This is contrary to the situation in Java where rice crops have been overfer-tilized. As a result, rice is largely unresponsive to additional P and responsivenesstends to be closely related to expectations based on soil P mapping.

Soils of Java and Lampung

Soil types are one of the elements of soil mapping that are used as a tool for producingfertilizer recommendations. The dominant soil types of Java and Sumatra differ; thus,the two regions need to be discussed separately (Table 1).

Alluvials and Latosols are the main soil types in Java used for lowland ricecropping. Under native vegetation, lowland soils in Java are relatively rich in nutri-ents because Java has a history of greater volcanic activity than the outer islands,especially eastern Indonesia (Amien 1997). The Latosols and Ultisols in the uplandsof Java have comparatively high native fertility and are highly weathered, acidic (pH4 to 5), low in most nutrients (Muljadi 1997), and susceptible to erosion even whenterraced (Amien 1997). However, with good fertilizer management, the uplands areproductive (Sri Adiningsih et al 1991).


Table 1. Area and distribution of soils in Indonesia(000 ha).

Soil type Java Sumatra

Red yellow podzolic (Ultisols) 325 15,950Organosols (Histosols) 25 6,781Alluvials (Entisols) 2,550 6,238Latosols (Inceptisols/Oxisols) 2,831 6,788Mediterranean (Alfisols) 1,625 –Andosols (Inceptisols) 844 2,725Podsols (Spodosols) – 931Regosols (Entisols/Inceptisols) 1,431 831Grumusols (Vertisols) 1,481 –Renzina (Mollisols) 38 394Complexes (mostly Ultisols) 2,069 6,725Total 13,219 47,363

Source: Muljadi and Soepraptohardjo (1975).

In Sumatra, upland cropping constitutes the bulk of agricultural land use (88%)and red-yellow podzolic soils are the dominant soil type (Santoso 1991), particularlyin areas newly opened for agriculture through the transmigration programs. The ma-jor region of agricultural development is the southern-most province of Lampung.The soils of the newly developed regions of Lampung are high in kaolin (>30%) withlow organic matter (<1%), low water-holding capacity, low P, N, K, S, Mg, Ca, andZn status, and high P fixation capacity (Sudjadi 1984, Prasetyo et al 1997). Red-yellow podzolic soils also tend to be acidic (pH 4 to 5) and this can lead to problemssuch as Ca and Mg deficiencies, readily leached K, high P, S, and Mo sorption, andexcessive hydrogen ions. In addition to acidity problems, plant growth can be limitedin these same soils through high Fe concentrations occurring within a few centime-ters of the soil surface (Prasetyo et al 1997) and/or high Al saturation (Santoso 1991).The problems associated with the use of red-yellow podzolic soils for upland riceproduction are destined to be of increasing significance as previously uncultivatedland is made available to farmers.

Organosols, alluvials, and Latosols are also prevalent particularly in the coastalswamp lands in Sumatra that have been developed as part of the transmigration pro-gram (Widjaja-Adhi et al 1996). The alluvial soils in Sumatra contain pyrites and areacid sulfate soils. The extent of the various soil types in the outer island’s coastalswamp lands was mapped in 1991-92 on a 1:500,000 scale using land unit maps andsoil maps produced by CSAR. Areas with deep peats (>3 m) or high susceptibility toerosion as determined by a set of distinct criteria are considered to be unsuitable forfood cropping (Widjaja-Adhi and Karama 1994).

Organosols have high N and organic carbon (OC) contents, but productivity onthese soils is poor due to subsidence after drainage, the slow release of OC and N, andhigh susceptibility to erosion through poor soil structure and slow water infiltrationrates (Sudjadi 1984). Efforts to overcome these limitations have focused on encour-

172 Clough et al

aging small landholders to farm together as a group and controlling water movementacross a series of farms (Widjaja-Adhi and Karama 1994). Newly cleared Organosolsintended for the transmigration program tend to have less nutrients than the peats thathave been farmed by spontaneous migrants for decades (Ruddle 1987).

Soil mapping has helped identify the main soil types in Java and Sumatra andhas highlighted the differences between the two regions. Based on mapping and re-lated field trials, fertilizer recommendations have been developed that account for thedifferences in soil types.

Development of phosphorus and potassium mapsand fertilizer recommendations for lowlands

Developing soil phosphorus maps and fertilizer recommendationsBefore 1972, P fertilizer recommendations did not take into account the effects ofdiversity of soil types, chemical properties of soils, or cropping histories. During the1960s, a blanket rate of 20 kg P ha–1 for both upland and lowland rice was recom-mended to all farmers throughout the country. Efforts to improve the specificity of Precommendations for particular regions and soil types have focused on soil mappingand relating those maps to results from fertilizer field trials in the lowlands. Much ofthe mapping work has been conducted by CSAR with some work being done throughcollaboration with overseas institutions.

Lowland soil P status mapping in Java began before 1970. In the first edition ofthe soil P maps for Java that were completed in 1971 (Widjaja-Adhi, personal com-munication), lowland areas were divided into two classes: responsive and nonrespon-sive to P fertilizer application. All soil P values for the first and following series ofmaps were determined by extraction in 25% HCl. Soils were deemed to be nonre-sponsive if extractable soil P was higher than 88 mg P kg–1 (20 mg P2O5 100 g–1).This cutoff point was based on findings that yield response to P fertilizer generallyoccurred only in soils where the P status was less than 88 mg P kg–1. Based on theinitial soil map, there were two recommendations: 0 and 20 kg P ha–1 crop–1.

The accuracy of the first map was limited because samples were taken onlyfrom research stations where field trials were also held. The relationship betweenfield response trials and soil P status at the research stations was extrapolated to farm-ers’ fields by assuming that soil P status and crop response were related to soil type.Thus, the first soil P status map was based on soil type.

The second edition of soil P maps for lowlands in Java was developed in 1974(Widjaja-Adhi, personal communication). These maps were also based on soil typebut with extensive analyses of soils for reserved P estimations taken from a widerange of locations. The third and fourth editions of the soil P maps of Java werecompleted in 1984 and 1992, respectively (Widjaja-Adhi, personal communication).Both the third and fourth editions divided the soils into three P status classes: high,medium, and low, which required 10 to 25 kg P ha–1 at varying frequencies (Table 2).The main change between the third and fourth editions of the P status maps was thatthe number of soil samples used to identify whether an area had low, medium, or high


Tabl

e 2

. Lo

wla

nd a

rea

(ha)

cla

ssifi

ed a

s lo

w,

med

ium

, an

d hi

gh s

oil

P a

nd K

sta

tus

for

13 I

ndon

esia

n pr

ovin

ces

(per

cent

age

of t

otal

are

a gi

ven

inpa

rent

hese

s).

Phos

phor

us s

tatu

sPo

tass

ium

sta

tus

Prov

ince

Low

Med

ium

Hig

hLo

wM

ediu

mH

igh

Wes

t Ja

va2

35

,62

1(1

9)

45

4,3

96

(37

)5

23

,34

8(4

3)

22

5,6

25

(19

)4

96

,25

0(4

1)

49

1,4

90

(41

)C

entr

al J

ava

12

3,4

39

(10

)6

58

,78

5(5

6)

39

7,1

20

(34

)1

75

,05

0(1

5)

33

0,0

00

(28

)6

74

,29

4(5

7)

East

Jav

a1

83

,50

0(1

5)

54

4,9

45

(43

)5

31

,47

5(4

2)

71

,87

5(

6)

34

5,6

25

(27

)8

42

,42

0(6

7)

Aceh

48

,22

4(1

6)

12

8,1

16

(43

)1

20

,81

8(4

1)

12

,07

1(

4)

56

,50

5(1

9)

22

8,5

82

(77

)N

orth

Sum

atra

53

,44

0(1

0)

30

1,5

98

(57

)1

75

,42

5(3

3)

10

,13

5(

2)

43

0,6

33

(81

)8

9,6

95

(17

)Ja

mbi

30

,47

0(1

2)

11

8,1

80

(45

)1

15

,83

1(4

4)

19

,59

5(

7)

13

9,9

35

(53

)1

04

,95

1(4

0)

Wes

t S

umat

ra3

7,3

89

(17

)9

5,9

83

(43

)9

1,7

93

(41

)5

0,3

98

(22

)1

10

,71

1(4

9)

64

,05

6(2

8)

Sou

th S

umat

ra1

45

,57

0(3

4)

25

1,9

81

(59

)3

2,3

15

( 8

)1

2,9

10

( 3

)2

61

,29

0(6

1)

15

5,6

66

(36

)La

mpu

ng1

7,7

07

( 8

)4

7,4

53

(22

)1

47

,92

2(6

9)

10

4,0

48

(49

)5

3,8

24

(25

)5

5,2

10

(26

)S

outh

Kal

iman

tan

14

5,8

29

(31

)1

64

,20

6(3

5)

15

5,1

86

(33

)6

6,2

52

(14

)2

61

,33

3(5

6)

13

7,6

36

(30

)S

outh

Sul

awes

i1

15

,44

8(2

0)

17

5,4

56

(30

)2

90

,11

6(5

0)

26

,66

9(

5)

89

,07

0(1

5)

46

5,2

81

( 8

0)

Bal

i1

,99

6(

2)

15

,52

1(1

7)

74

,05

4(8

1)

–(

0)

–(

0)

91

,57

1(1

00

)Pa

lau

Lom

boka

–(

0)

11

,65

2(1

0)

11

0,8

33

(90

)–

( 0

)–

( 0

)1

22

,48

5(1

00

)

a Reg

ion

in th

e N

usa

Teng

gara

Bar

at P

rovi

nce.

Low

soi

l P <

88 m

g kg

–1, m

ediu

m s

oil P

88

–17

6 m

g kg

–1, h

igh

soil

P >

17

6 m

g kg

–1. L

ow s

oil K

<1

66

mg

kg–1

, med

ium

soi

l K 1

66

–3

32

mg

kg–1

, hi

gh s

oil K

>3

32

mg

kg–1

. D

ata

sour

ced

from

CS

AR (

unpu

blis

hed)

.

174 Clough et al

soil P was increased. In 1998, the fourth edition of lowland maps for Java and theouter islands was digitized by CSAR.

The definition of low, medium, and high soil P was derived from P responsive-ness field trials conducted throughout the lowlands of Java (Table 3). The lack ofresponse to P fertilizer was demonstrated in 18 and 15 trials conducted by Sri Adiningsih(unpublished) over the 1987-88 wet season and 1988 dry season, respectively. Ricegrown in the wet season responded to P at only seven sites and rice grown in the dryseason responded at only two sites. The significant yield increases tended to be small(0.5 t ha–1) in terms of actual grain yields. Trials with rice on experimental farmslocated at Muara in West Java and Ngale in East Java in the wet and dry seasons alsoproduced no response to P applications (Miyake et al 1984).

The fourth edition (Widjaja-Adhi and Sri Adiningsih, personal communica-tion) of lowland soil maps for Java (scale 1:500,000) showed that most of the low-lands of Java were classified as having medium to high soil P status (Fig. 1). Thedominance of medium to high P status is because the lowlands have received P fertil-izer in excess of that required for rice crops for many years. Over the past 30 years, Phas been applied to irrigated and rainfed lowland rice fields at an average of 20 kg Pha–1 crop–1. Subsequently, extractable soil P has increased as P has accumulated in thesoil and P fertilizer requirements in 1999 are low (Table 2). Mapping data and resultsfrom lowland rice trials indicate that P fertilizer need only be applied to most lowlandrice in Java at rates equivalent to the amount of P removed in the grain. High-yieldingrice cultivars may produce between 5 and 8 t grain ha–1 (Hermanto 1995); thus, Pfertilizer needs to be applied at between 12.5 and 20 kg ha–1 crop–1 to avoid removingP from the soil. However, most lowland areas in Java can be cropped for severalseasons without a depletion of soil P resulting in a decline in grain yield.

Soil P maps (1:500,000) of the lowland areas of Lampung also show that largeareas have a high soil P status (Fig. 2), which has been extensively developed foragricultural use through the transmigration programs. However, using the same clas-sification system, other provinces such as South Sumatra are mainly classified ashaving low to medium soil P status (Table 2). Lowland regions on most of the otherislands that have been mapped are deemed to have low to medium soil P status. Ex-ceptions are the soils of Bali and Lombok (Fig. 3), which have medium to high soil Pbecause the islands are volcanic.

Table 3. Phosphorus fertilizer recommendationsfor lowland rice grown in Java based on soil Pstatus as determined by 25% HCl extraction(Hermanto 1995).

Extractable P Recommended P rate(mg kg–1) (kg ha–1)

<88 20–25 every season88–176 15 every second season>176 10 every fourth season


Fig.

1. E

xtra

ctab

le s

oil P

map

(25%

HC

l ext

ract

ion)

of t

he lo

wla

nds

of W

est J

ava

(Jaw

a B

arat

) pro

duce

d in

1998 (1:5

00,0

00)

(<88, 88–1

76, an

d >176 m

g P

kg–

1 a

re in

dica

ted

by r

ed, ye

llow

, an

d gr

een

shad

ings

, re

spec

tive

ly).

176 Clough et al

Fig. 2. Extractable soil P map (25% HCl extraction) of the lowlands of Lampung Province in Sumatraproduced in 1998 (1:500,000) (<88, 88–176, and >176 mg P kg–1 are indicated by red, yellow, andgreen shadings, respectively).


Fig.

3.

Extr

acta

ble

soil

P m

ap (

25%

HC

l ex

trac

tion

) of

the

low

land

s of

Pal

au L

ombo

k in

the

pro

vinc

e of

Nus

aTe

ngga

ra B

arat

pro

duce

d in

1993 (

1:2

50,0

00)

(88–1

76 a

nd >

176 m

g P

kg–

1 a

re in

dica

ted

by y

ello

w a

nd g

reen

shad

ings

, re

spec

tive

ly).

178 Clough et al

Benefits and limitations of soil phosphorus mappingIn the 1970s, an undeveloped upland and lowland area of Sitiung (Mimpi Plain) inWest Sumatra was forwarded as a potential area for settlement of transmigrants andfor upland research coordinated as part of the Benchmark Soils Project conducted bythe University of Hawaii (Buurman and Sukardi 1980). The area was mapped at amoderate resolution (1:40,000) for soil type by CSAR by taking soil samples frommore than 50 locations scattered throughout the area in 1976-77 (Soil Research Insti-tute 1979). The mapping project demonstrated the benefits of mapping for agricul-tural development policy and soil classification, and the limitations of low-resolutionmapping conducted by CSAR.

Some of the sites in Sitiung were classified in detail with samples being takento a depth of 180 cm. This detail enabled several soils in the region to be reclassifiedamong the classes brown tropical forest, Latosol, and podzolic (Buurman and Sukardi1980). Soil profiles from 88 sites in the same region were later used to map topsoilproperties including KCl-extractable Al, extractable P (25% HCl), sand and clay con-tents, total P, and pH, and 109 scattered samples were used to map variation in siltcontent. Actual soil property values were translated into maps by krigging to give 268points within the sample area (Trangmar et al 1984). The soil P status maps fromSitiung showed that extractable soil P was low to medium throughout the region (mainly40–160 mg P kg–1) with high P sorption capacity compared with other tropical Oxisolsand Ultisols. Mapping showed that native extractable soil P in uplands and lowlandswas correlated positively with silt content (r = 0.52**) and negatively with sand con-tent (r = –0.55**) at 0 to 15-cm depth.

The relationship between soil P and soil texture in Sitiung demonstrates thatCSAR’s original soil P recommendation maps completed in 1971 (Widjaja-Adhi,personal communication), which were based on soil type rather than directly measur-ing soil P in each field, were well founded. However, the variation in soil P in Sitiungshows that a weakness in mapping conducted by CSAR may be in the resolutionsince the maps are only 1:250,000–1:500,000 scale. This low resolution may explainwhy not all field trials gave the responses as expected based on the soil maps. Forexample, P response trials conducted near Indramayu and Subang in 1987-88 re-sponded to P application despite the area being classified as having high P status (SriAdiningsih, unpublished).

Good correlation between actual yield and expected yield based on the soilnutrient status maps may also be limited because the maps do not distinguish irri-gated, rainfed, or tidal lowlands. All the lowland types are presented on one map andrecommendations that follow from the mapping exercises and field trials apply to alllowland rice-growing conditions. This may limit the usefulness of the maps as a toolfor formulating P recommendations since average grain yields and farmers’ aversionto the risks involved in investing in fertilizers differ between cropping systems.

The fourth edition of soil P maps shows that much of the lowlands of Lampungin Southern Sumatra have a high soil P status. This was confirmed by a soil survey of32 irrigated lowland sites conducted in 1998 before beginning field trials in a pro-gram known as the Acid Soils Project (Kasno, unpublished). The survey showed that,


based on the low, medium, and high classifications used in Java, only 11 lowland siteswere likely to respond to P applications. One of the sites expected to be responsive,Braja Mas, was selected for the Acid Soils Project and was transplanted to rice in the1998-99 wet season. The soil P for the site was 55 mg P kg–1 (25% HCl extraction).As expected, the rice crop gave a positive response to P with 90% maximum grainyield achieved with 52 kg P ha–1. The amount of P required is close to that recom-mended by CSAR for all lowlands in Sumatra (Table 4). The recommendations arehigher for lowlands in Sumatra than for those in Java because of the need to over-come P fixation in some soils and the soils’ inherent low fertility. Rice grown infarmers’ fields is responsive to P applications and the potential exists to increase soilP status to the point where P applications are no longer required as is the case in mostlowlands of Java (Fig. 1).

In Lampung, particularly on the red-yellow podzolic soils, high P fixation ca-pacity limits the relevance of soil P status to P recommendations. This was shown inan irrigated field trial conducted at Pringsewu in Lampung as part of the Acid SoilsProject in the 1998-99 wet season. The Pringsewu site had very high soil P (>300 mgP kg–1) yet responded to P application, producing 90% maximum rice grain yield (4.2t ha–1) with 59 kg P ha–1.

An additional detrimental effect of lowland soils with high P-fixing capacity istheir tendency to fix any P that is not used in the year of application. CSAR recom-mendations for lowlands in Java are to apply P only every fourth season if soil P isclassified as high. However, two irrigated lowland trials in Lampung conducted dur-ing the Acid Soils Project showed that P applied in excess of crop requirements wasnot extractable by 25% HCl at the end of the growing season. This type of field resulthas led to the recommendation that P be applied every season in Sumatran lowlands.

Validation of soil phosphorus mappingIn 1994-95, the maps of East, Central, and West Java compiled by CSAR were usedas a baseline for predicting the change in status of extractable soil P after severalcropping seasons (Pandutama 1996). Soils for the study were collected from severallowland fields throughout Java covering the main soil types present on the island:

Table 4. Recommendations for P application onlowlands of Sumatra.

Extractable P Recommended P ratea

(mg kg–1) (kg ha–1)

>88 066–88 3344–65 6622–43 99<22 132

aRecommendations are for each cropping season. Datasourced from CSAR (unpublished).

180 Clough et al

Latosol, Regosol, alluvial, and Grumusol (Oxisol, Entisol, Inceptisol, Vertisol). Thelocation of each sampling was based on the fourth edition of soil P maps produced byCSAR in 1992 to ensure that soils with low, medium, and high P status were includedin the study. Soil P (25% HCl extraction) for the ten sites selected ranged from 83 to633 mg P kg–1 (median = 258 mg P kg–1). A series of rice crops were grown in eachsoil with and without P fertilizer. At each site, the change in soil P over the experi-mental cropping period was related in a model to clay content, organic carbon con-tent, physical clay activity, initial soil P status (25% HCl extraction), and P applica-tion rate. Soil parameters in the model had been selected from 12 potentially influen-tial soil characteristics using multiple regressions. The model was then used to pre-dict the decline in soil P after rice cropping without P application given a particularinitial soil P value for each soil type.

Predictions by Pandutama (1996) confirmed and refined what CSAR had foundbased on its nutrient status mapping and fertilizer field trials. Pandutama’s (1996)estimates of P requirements based on mapped P status and rates of decline with crop-ping predicted that lower rates (10 kg P ha–1) could be applied to achieve yields of upto 4 t ha–1 than the rates (15 to 25 kg P ha–1) recommended by CSAR for soils withmedium to high P status. Although the target yield is less than the maximum potentialyield of modern varieties, the yield is realistic for Javanese farmers.

Developing soil potassium maps and fertilizer recommendationsSoil mapping for extractable K (25% HCl extraction) in the lowlands was conductedusing techniques similar to soil P mapping, that is, information was collected overseveral years and recommendations were created based on results of soil samplingand field trials, although K recommendations are not as well developed as P recom-mendations.

In 1989, an extractable soil K status map of Java, including Madura, was madebased on 600 soil samples and some field trials. Like the soil P maps of Java, soil Kmaps divide the lowlands into three categories: high, medium, and low. The 1989map indicated that 40% of lowland soils had low K status (<20 mg K kg–1 in 25%HCl extraction) (Didi Ardi et al 1989). In 1996, less than 13% of lowland soils on theisland of Java required K fertilizer (Karama et al 1998).

In Java, the reduction in the amount of land with low K is due to farmers apply-ing K at about 3 to 4 kg ha–1 since 1978 (Sri Adiningsih et al 1990, O’Brien et al1990) and farmers adopting the practice of incorporating rice residues, which cancontain 40% to 60% of the K taken up by the crop (Dobermann et al 1996a). Conse-quently, many lowland rice crops in Java are no longer responsive to K application(Sri Adiningsih et al 1990, Sri Rochayati et al 1990) and many of Java’s lowland soilsare classified as high (Fig. 4). Lack of response to K in the lowlands of Java indicatesthat K fertilizers need not be applied in many regions, as shown in the soil K maps.The areas of lowland Java that are low in K also tend to be low in soil P, as shown inthe P and K maps of Java (Figs. 1 and 4). Some sites, however, are still responsive toK as demonstrated by K trials at the Jakenan Experimental Station in Central Javawith rainfed rice (Wihardjaka et al 1998). The two initial consecutive trials at Jakenan


Fig.

4. E

xtra

ctab

le s

oil K

map

of t

he lo

wla

nds

of W

est

Java

(Ja

wa

Bar

at)

prod

uced

in 1

998

(1:5

00,0

00)

(<83, 8

3–

166, an

d >166 m

g K

kg–

1 a

re in

dica

ted

by r

ed, ye

llow

, an

d gr

een

shad

ings

, re

spec

tive

ly).

182 Clough et al

gave a positive yield response with application of 75 kg K ha–1. In the third, fourth,and fifth trials, there was no increase in grain yields due to either increases in soil Kstatus (third and fifth trials) or low in-crop rainfall (fourth trial).

The prevalence of high soil K status in Java needs to be considered along withthe K balance of the soil when formulating K fertilizer recommendations. A positivesoil K balance was shown to occur when K was applied to some treatments in Jakenanrainfed rice trials. In the treatments at Jakenan where N and P were added without K,the K balance was negative, whereas adding NPK (120:18:75) gave an increase insoil K (Wihardjaka et al 1998). Trials with irrigated rice in Java (Dobermann et al1996a), however, showed that K uptake was greater than the amount applied, thuscreating a negative K balance in the soil at the end of the season. Cropping systemswith a negative K balance are unsustainable in the long term and K fertilizers need tobe applied to reduce the risk of K deficiencies. Thus, there is only one recommendedrate for the whole of Java: 26 kg K ha–1 (Karama et al 1998).

Lowland rice has been shown to be more responsive to fertilizer K in Sumatrathan in Java. This reflects the fact that low-K areas are more prevalent in Lampung(Fig. 5). Unlike in Java, however, soil P maps are not related to soil K maps. Resultsfrom soil K mapping in Lampung are supported by a soil survey conducted in 1998for the Acid Soils Project, which found that 25 of 32 lowland sites had low soil K(<50 mg K kg–1) as determined by 25% HCl extraction. Surveys in West and SouthSumatra showed that only 10% of the lowland soils had high soil K (>166 mgK kg–1) (Karama et al 1998). Despite the need for K fertilizer to be applied in thelowlands of Sumatra, no K fertilizer recommendations specific to that region havebeen developed.

Developing K recommendations is hampered by the difficulty in relating soil Kextracted by 1M NH4Oac to grain yields. Dobermann et al (1996b) concluded thatpredicting crop K uptake in irrigated rice using a static soil test was not practicalbecause too many soil properties (extractable K, Ca, and Mg combined, CEC, or-ganic matter, clay content) were required to give accurate results. K uptake in irri-gated lowland rice has been shown to be related to soil K (r2 = 0.82) as measuredusing mixed-bed ion exchange resin capsules (Dobermann et al 1996b). Ion exchangeresins have the advantage of being able to measure the amount of K that can be ex-tracted from a soil over time rather than measuring K in solution at a particular mo-ment. Dynamic measurements of K are required for intensive cropping systems andwhere the soils have K-fixing properties.

Data used to test the validity of predicting total K uptake at harvest in lowlandrice were derived from NPK trials conducted at 11 sites in five countries includingIndonesia with K as the only limiting factor (Dobermann et al 1996b). K fertilizerrates varied between sites from 25 to 100 kg K ha–1, soil pH ranged from 5.7 to 8.5,and clay contents ranged from 25% to 57%. This same method could possibly be usedto predict K uptake by rainfed lowland rice in locations within Java and Lampung.


Fig. 5. Extractable soil K map of the lowlands of Lampung Province in Sumatra produced in 1998(1:500,000) (<83, 83–166, and >166 mg K kg–1 are indicated by red, yellow, and green shadings,respectively).

184 Clough et al

Phosphorus and potassium recommendations for uplands

Developing phosphorus recommendationsApart from isolated exercises such as the mapping of Sitiung in the 1970s, very littlesoil mapping of uplands has been conducted. The lack of work in upland mapping isprimarily due to the uplands making only a small contribution to national rice pro-duction compared with the lowlands, especially in Java. Phosphorus recommenda-tions are therefore based solely upon responses of rice grown in field or pot trials andtend to still be broad.

Recommendations for upland rice and other food crops are 20 to 40 kg P ha–1 inthe first crop and 20 kg P ha–1 for each crop thereafter. The difference between actualand recommended P fertilizer rates applied to upland rice in Java is small. As such,the upland area that is low in soil P is less than 20% for Java and getting smaller. InLampung, the amount of upland rice-growing area that has low soil P is also declin-ing; however, responsive sites are still prevalent. A survey of 11 upland sites con-ducted when the Acid Soils Project began showed that a response to P could be ex-pected at 10 of the sites (Kasno, unpublished).

The upper limits of the recommendations are constrained by farmers’ potentialto invest in P fertilizer in a rainfed environment where risk of crop failure is com-pounded by the presence of other nutritional deficiencies. The need for a holisticapproach to fertilizer management in upland rice cropping was shown in a trial con-ducted in the 1998-99 wet season at the Taman Bogo Experimental Farm as part ofthe Acid Soils Project in Lampung. The trial at Taman Bogo gave a positive responseto SP-36 (16% P as superphosphate) applied at 80 kg P ha–1, but only when applied inconjunction with >80 kg N ha–1 (Clough, unpublished).

The validity of these broad P recommendations for upland crops was exempli-fied, however, by trials with upland rice on newly formed bench terraces with extract-able soil P (Bray I) of 6 mg P kg–1. Grain yields in upland rice increased upon appli-cations of 0 to 40 kg P ha–1 in three consecutive years (Schmidt et al 1990). Meanyield across all treatments was 1.88, 2.40, and 2.59 t ha–1 in 1984, 1985, and 1986,respectively. Within treatments, the average yield across the three years was 0.35,2.45, 2.89, and 3.46 t ha–1 for applications of 0, 10, 20, and 40 kg P ha–1, respectively.In low P-fixing conditions, such as those presented by Schmidt et al (1990), soil Pmapping in the uplands of Java may be an effective means of improving P recommen-dations. In Lampung, however, where the uplands are dominated by P-fixing red-yellow podzolics, soil mapping may be of limited value.

Limitations of using soil phosphorus mappingfor developing fertilizer recommendations due to P fixationP fixation capacity is a significant issue for P fertilizer management in the red-yellowpodzolic soils of Lampung and other provinces of Sumatra. High P fixation capacityof soils is more likely to occur when several of the following factors are present(Brady and Weil 1999):


● acidic pH (< 6)● low organic matter content● high clay content, particularly kaolinite, gibbsite, or goethite● high concentrations of Fe, Al, or Mn ions in soil solutionThe variability in response of rice to P applications because of P fixation in red-

yellow podzolic soils has been demonstrated in several P trials. P response trials withupland rice in Lampung have shown an increase in grain yield with P application atsome sites with relatively low soil P, particularly in newly cleared areas (Effendi et al1982). However, a two-year trial with upland rice in Pekalongan, Lampung, gave noresponse to P (Makarim 1990), whereas upland rice grown on a clayey Typic Paleudultproduced 90–100% maximum grain yields at 20 kg P ha–1 (Palmer and Sudjadi 1984).A trial on a newly cleared site in Central Lampung showed that grain yields could bedoubled by applying P fertilizer at rates similar to those recommended for irrigatedrice in Java (Effendi et al 1982).

The dual effects of extractable soil P and P sorption capacity are exemplified intwo upland P trials conducted in Lampung during the 1998-99 wet season as part ofthe Acid Soils Project. Site selection was based on soil P maps produced by CSARand 1998 soil surveys. Phosphorus was applied at four rates, the maximum rate beingdetermined through the soils’ P sorption curve to be adequate to give 0.02 µg mL–1

soil P in solution. Rice crops grown at both sites gave a positive response to SP-36application (P <0.005). The magnitude of the grain yield response was related to thesites’ soil P (25% HCl extraction) before P treatment and P sorption capacity. Soil Pwas 118 and 191 mg P kg–1 and maximum P fertilizer application was 327 and 96 kgP ha–1 at Buyut Udik and Jagang, respectively. The amount of P fertilizer required togive 90% maximum grain yield was 95 kg ha–1 at Buyut Udik and 55 kg ha–1 atJagang. This means that Jagang, the site with relatively high initial soil P and low Psorption, required less P fertilizer for the rice to achieve maximum grain yield thanBuyut Udik. The diversity of results from response trials reflects the range of extract-able soil P values and P sorption capacity found at upland sites in Sumatra.

A tactic promoted to reduce the problems of P fixation is to apply rock phos-phate (RP) instead of readily available SP-36 or triple superphosphate (TSP). A re-view of P response studies in upland rice (Partohardjono and Sri Adiningsih 1991)concluded that, although the initial grain yields with RP were lower than with TSP,over several seasons five RP fertilizers from Indonesia and Christmas Island werejust as effective as TSP. In 1998, as part of the Acid Soils Project, an upland trial wasestablished to determine whether a blend of readily available phosphate and rockphosphate could reduce P loss through fixation while maintaining yields in the firstseason. The trial compared locally derived RP (Ciamis), SP-36, and Prolong (sup-plied by Pivot Ltd.). Prolong is a combination of superphosphate and reactive rockphosphate (P:S:Ca = 9.5:15:23), which is designed to provide a readily available sourceof P to crops and a source that is slowly released over one or more seasons. Phospho-rus was applied at 0, 20, 40, and 80 kg P ha–1. As expected from a low-P-fixing sitewith medium soil P status, all P sources gave a significant increase in grain yield (P =0.005) with 90% maximum grain yield being achieved with 40 kg P ha–1. The re-

186 Clough et al

sponse of the crop to the three P sources differed with Prolong giving significantlyhigher grain yields than the other two sources (P = 0.007). The rice crop achieved90% maximum grain yield with 28 kg Prolong-P ha–1 compared to requiring about 60kg P ha–1 applied as SP-36.

However, these findings that show that commonly available P sources are equallyas effective over time at increasing yields must be judged relative to the cost of eachP source. Economically, RP is more efficient than SP-36 with rice farmers payingabout Rp 500 (US$0.07) kg–1 of RP fertilizer (Rp 60 per kg P) and about Rp 2,500(US$0.38) kg–1 of SP-36 fertilizer (Rp 400 per kg P) in Lampung in July 1999. Atpresent, rice farmers are willing to pay a maximum of Rp 1,500 (US$0.23) kg–1 offertilizer.

Developing potassium recommendationsAs stated previously, no intensive effort has been made to create soil K maps from theuplands of Lampung or Java; consequently, K recommendations are broad.

Upland rice grown in Lampung has been shown to be responsive to K. Resultsfrom field trials were supported by a soil survey conducted in 1998 for the Acid SoilsProject, which found that 9 of 11 upland sites had low soil K (<50 mg K kg–1) asdetermined by 25% HCl extraction. The prevalence of areas with low soil K in theuplands of Lampung is partially due to the dominance of acid soils. Under acidifica-tion processes that occur in upland areas of Lampung, adsorbed K+ is replaced by H+

and Al3+. Subsequently, K+ is easily leached down the soil profile. This situation isaccentuated by the availability of K being lower in sandy or acid soils or where theclay mineralogy is dominated by kaolinite, as is the case in upland Lampung. HighCa and Mg in a soil, also present in uplands, can further reduce K uptake by plants.

All treatments in the lowland and upland trials at the Taman Bogo Experimen-tal Farm conducted as part of the Acid Soils Project received a basal application ofKCl at 60 kg K ha–1. The mean K concentration of the upland rice grain was slightlylower (0.15%) than the mean value (0.21%) given for four rainfed K trials at theJakenan Experimental Station (Wihardjaka et al 1998). The K balance was positivewith about 27 kg ha–1 of soil K available at the site in addition to applied K. Total Kcontent in the grain and straw was about 42 kg ha–1 where no N or P was applied and56 to 73 kg K ha–1 where the crop was given 80 kg P ha–1 and 40–160 kg N ha–1.Other upland rice trials in Lampung have found the K balance to be negative (Gill andKamprath 1990) with K recovered being greater than 100%. Recoveries greater than100% in the upland rice were attributed to applied K stimulating root growth andconsequently increasing uptake of native K. Additional K uptake even in nondeficientrice crops has been shown to be beneficial for increasing grain yields through beingaccompanied by a reduction in the incidence of disease (Gill and Kamprath 1990) oriron toxicity (Ismunadji 1990).

In upland Sumatra, where soils are inherently low in K, the application ratesneed to be increased to match K removal and possibly assist in alleviating the detri-mental effects of iron toxicity. General K fertilizer recommendations for upland cropswere developed by CSAR based on field trials in Lampung and Java. Based on field


trials, K recommendations for upland rice and maize are 66–100 kg K ha–1 (SriAdiningsih et al 1991), which are far higher than the rates recommended for irrigatedlowland and higher than the amounts applied by upland farmers. The higher recom-mended rates reflect the poor native fertility of upland soils compared with lowlandsoils.

Conclusions

Mapping soil type and P and K status is a long-term task that has been carried out inIndonesia continuously for more than 30 years. Over that 30-year period, the mapshave become more detailed by dividing provinces into three instead of two nutrientstatus categories and maps have been made more accurate by using large sample sizesto categorize regions. Mapping P and K status in lowland Java has been completedand efforts now focus on mapping lowlands in the outer islands. Extensive progresshas been made in mapping Sumatra, Bali, and Lombok. Comparisons between P andK maps show that low P is usually accompanied by low K except in some regionssuch as Lampung.

The combination of mapping and field trials has led to the development of Pfertilizer recommendations based on soil P for the lowlands of Java. To this end, themaps for P status in Java are adequate despite the low resolution. Adequacy primarilystems from the fact that overall P status in Java is high due to intensive P fertilizerapplication. P fertilizer recommendations in Java may be improved by superimposinga rice-cropping system map (irrigated, rainfed, tidal) over the P status maps since Pfertilizer has not been evenly applied across the three production systems (Widjaja-Adhi, personal communication). The P recommendations for Java also appear to beapplicable in other regions such as Bali and Sumatra where P sorption is not a signifi-cant issue. Together, mapping and field trials show that P fertilizer is only required inlimited amounts in the lowlands of Java and at rates that are declining in Lampung.Improvements in P recommendations may be acheived in some areas of Lampung bymapping at a higher resolution where current maps show that soil P status varieswithin small (20 km2) lowland rice production areas.

Potassium recommendations are less specific than P recommendations. The map-ping, however, is comprehensive, especially in Java, and K status-specific recom-mendations could be developed from field trials conducted in the region. Trials, map-ping, and surveys show that the benefits of K fertilizer applications are underesti-mated by farmers in terms of both direct yield increases and the reduced risk of irontoxicity.

Mapping in upland regions is less developed than in lowland regions; this isreflected by less specific P and K fertilizer recommendations being available to up-land rice farmers. Fertilizer trials in the uplands show that P and K applications aremore likely to be beneficial there than in the lowlands. Phosphorus has the addedproblem of P sorption in the red-yellow podzolic soils that are prevalent in Lampung.Recommendations specific to the uplands of Lampung need to take this into accountby using soil P status and the soils’ sorption capacity in models.

188 Clough et al

ReferencesAmien I. 1997. Land resource analysis for agriculture development. Indones. Agric. Res. Dev.

J. 19:1-12.Brady NC, Weil RR. 1999. Soil phosphorus and potassium. In: The nature and properties of

soils. Upper Saddle River, NJ (USA): Prentice-Hall Inc. p 540-584.Buurman P, Sukardi. 1980. Brown soils, latosols or podzolics? In: Buurman P, editor. Red soils

in Indonesia. Wageningen (Netherlands): Centre for Agricultural Publishing and Docu-mentation. p 93-104.

Didi Ardi S, Supardi AK, Aldjabri M, Supartini J, Adiningsih S, Sudjadi M. 1989. The status ofpotassium in lowland soils in Java and Madura. Bogor (Indonesia): Centre for Soil Re-search.

Dobermann A, Cassman KG, Sta. Cruz PC, Adviento MA, Pampolino MF. 1996a. Fertilizerinputs, nutrient balance, and soil nutrient-supplying power in intensive, irrigated ricesystems. II. Effective soil K-supplying capacity. Nutr. Cycl. Agroecosyst. 46:11-21.

Dobermann A, Sta. Cruz PC, Cassman KG. 1996b. Fertilizer inputs, nutrient balance, and soilnutrient-supplying power in intensive, irrigated rice systems. I. Potassium uptake and Kbalance. Nutr. Cycl. Agroecosyst. 46:1-10.

Effendi S, Ismail IG, McIntosh JL. 1982. Cropping systems research in Indonesia. In: Reportof a workshop on cropping systems research in Asia. Manila (Philippines): InternationalRice Research Institute.

Gill DW, Kamprath EJ. 1990. Potassium uptake and recovery by an upland rice-soybean rota-tion on an oxisol. Agron. J. 82:329-333.

Hermanto MS. 1995. Rice production technology supporting sustainable rice self-sufficiencyin Indonesia. Bogor (Indonesia): Central Research Institute for Food Crops. p 62.

Ismunadji M. 1990. N, P, K and S status of food crops in islands outside Java. In: Blair G,Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice cropping systemsin Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for InternationalAgricultural Research (ACIAR) Proceedings No. 29. Canberra (Australia): ACIAR.p 83-86.

Karama AS, Fagi AM, Rochayati S. 1998. Current use of and requirement for nutrients forsustainable food crops production in Indonesia. In: Johnston AE, Syers JK, editors. Nu-trient management for sustainable crop production in Asia. Wallingford (UK): CABInternational. p 291-305.

Makarim AK. 1990. S research on upland crops in Indonesia. In: Blair G, Lefroy R, editors.Sulfur fertilizer policy for lowland and upland rice cropping systems in Indonesia, Jakarta,Indonesia, 18-20 July 1989. Australian Centre for International Agricultural Research(ACIAR) Proceedings No. 29. Canberra (Australia): ACIAR. p 91-94.

Miyake M, Ismunadji M, Zulkarnaini I, Roechan. 1984. Phosphate response of rice in Indone-sian paddy fields. Tech. Bull. Trop. Agric. Res. Cent. 17:1-78.

Muljadi D. 1997. Application of phosphate rock fertilizer based on soil and climatic conditionsin sustainable Indonesian agriculture. Indones. Agric. Res. Dev. J. 19:35-42.

Muljadi D, Soepraptohardjo M. 1975. Masalah data luas dan penyebaran tanah-tanah kritis. In:Simposium Pencegahan dan pemulihan tanah-tanah kritis dalam rangka pengembanganwilayah, Jakarta, Indonesia.


O’Brien DT, Blair G, Lefroy R. 1990. The fertilizer subsidy and fertilizer use in Indonesia. In:Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and upland rice croppingsystems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centre for Inter-national Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Australia):ACIAR. p 42-50.

Palmer B, Sudjadi M. 1984. Efficient use of fertilizer in upland cropping systems: an experi-mental approach. In: Bay-Petersen J, editor. Ecology and management of problem soilsin Asia. Food and Fertilizer Technology Center for the Asian and Pacific Region, No.27, Bangkok, Thailand. p 293-302.

Pandutama MH. 1996. Predicting change in residual extractable P for site-specific manage-ment in paddy soils in Indonesia. PhD dissertation. University of Hawaii, Hawaii.

Partohardjono S, Sri Adiningsih J. 1991. Response of food crops to phosphate rocks. Indones.Agric. Res. Dev. J. 13:46-57.

Prasetyo BH, Sulaeman, Sri Mulyani N. 1997. Red-yellow soils from Kotabumi, Lampung:their characteristics, classification, and utilization. Indones. J. Crop Sci. 12:37-45.

Ruddle K. 1987. The impact of wetland reclamation. In: Wolman MG, Fournier FGA, editors.Land transformation in agriculture. SCOPE Series No. 32. Paris (France): John Wileyand Sons. p 171-201.

Santoso D. 1991. Agricultural land of Indonesia. Indones. Agric. Res. Dev. J. 13:33-36.Schmidt DR, Casler MD, Saefuddin A. 1990. Crop response to lime and phosphorus on new

bench terraces in the tropics. Agron. J. 82:333-337.Soil Research Institute. 1979. Soil survey of the irrigation project area, Sitiung-S. Jujuhan

(West Sumatra). Ministry of Agriculture, Agency for Agricultural Research and Devel-opment, Bogor, Indonesia.

Sri Adiningsih J, Santoso D, Sudjadi M. 1990. The status of N, P, K and S of lowland rice soilsin Java. In: Blair G, Lefroy R, editors. Sulfur fertilizer policy for lowland and uplandrice cropping systems in Indonesia, Jakarta, Indonesia, 18-20 July 1989. Australian Centrefor International Agricultural Research (ACIAR) Proceedings No. 29. Canberra (Aus-tralia): ACIAR. p 68-76.

Sri Adiningsih J, Semali A, Effendi S, Hadiwigeno S. 1991. Resources and problems associ-ated with the development of upland areas in Indonesia. In: Blair G, Lefroy R, editors.Technologies for sustainable agriculture on marginal uplands in Southeast Asia. Ter-nate, Cavite, Philippines, 10-14 Dec 1990. Australian Centre for International Agricul-tural Research Proceedings, No. 33, Canberra, Australia. p 45-54.

Sri Rochayati, Muljadi D, Sri Adiningsih J. 1990. Penelitian efisiensi penggunaan pupuk dilahan sawah. In: Prosiding Lokarkya nasional efisiensi penggunaan pupuk. PusatPenelitian Tanah dan Agroklimat, Cisarua. p 107-143.

Sudjadi M. 1984. Problem soils in Indonesia and their management. In: Bay-Petersen J, editor.Ecology and management of problem soils in Asia. Bangkok (Thailand): Food and Fer-tilizer Technology Center for the Asian and Pacific Region, No. 27. p 58-73.

Trangmar BB, Yost RS, Sudjadi M, Soekardi M, Uehara G. 1984. Regional variation of se-lected topsoil properties in Sitiung, West Sumatra, Indonesia. Hawaii Institute of Tropi-cal Agriculture and Human Resources, University of Hawaii.

Widjaja-Adhi IPG, Karama AS. 1994. Problems related with the sustainable development ofcoastal plains in Indonesia: research needs and related priorities. Indones. Agric. Res.Dev. J. 16:37-47.

190 Clough et al

Widjaja-Adhi IPG, Suriadikarta DA, Karama AS. 1996. Reclamation and management of salt-affected acid sulphate soils for agricultural development. Indones. Agric. Res. Dev. J.18:37-43.

Wihardjaka A, Kirk GJD, Abdulrachman S, Mamaril CP. 1998. Potassium balances in rainfedlowland rice on a light-textured soil. In: Ladha JK, Wade L, Dobermann A, Reichardt W,Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient managementresearch. Proceedings of the International Workshop on Nutrient Research in RainfedLowlands, Ubon Ratchathani, Thailand, 12-15 Oct 1998. Manila (Philippines): Interna-tional Rice Research Institute. p 127-137.

NotesAuthors’ addresses: A. Clough, S. Fukai, School of Land and Food Sciences, The University of

Queensland, St. Lucia 4072, Australia; I.P.G. Widjaja-Adhi, J. Sri Adiningsih, A. Kasno,Centre for Soil and Agroclimate Research, Jalan Ir. H. Juanda 98, Bogor 15123, Indone-sia.

Acknowledgments: The authors wish to acknowledge the financial support provided by theAustralian Research Council (ARC) and Pivot Ltd. for the Acid Soils Project conductedby CSAR and The University of Queensland.


Planning and managing rice farming through environmental analysis 191

Environmental analysis provides insights into planning and managing ricefarming according to the prevailing conditions. This helps in developing situ-ation-specific technologies and in selecting areas suitable for the applicationof promising technologies, particularly in rainfed conditions. Studies carriedout in eastern India characterized and classified the rainfed rice area in dif-ferent rice ecosystems and subecosystems and elucidated the main prob-lems and opportunities for enhanced productivity in them. A detailed analy-sis of the principal rice subecosystems provided the periods and quantitiesof water surplus and deficit and moisture use and recharge patterns alongwith other climatic variables during the rice-cropping season.

This information was used for selecting the technological interventionsthat were considered suitable for managing rice farming in such areas. Theinterventions were compared with the farmers’ normal practices by monitor-ing crop performance at selected locations in two rice ecosystems.

Rice environmental analysis is carried out for various purposes. It provides usefulinformation on what the environments are, how much rice area there is, and howproduction is distributed across environments. It is supposed to enhance the resource-use efficiency and impact of technologies. It provides insights into planning and man-aging rice farming and selecting areas suitable for the application of promising tech-nologies.

This chapter reviews some of the environmental analysis that has been carriedout in the rainfed regions of eastern India. The first part presents the ecosystem char-acterization from the mega to micro level and the classification of rice areas intobroad ecosystem/subecosystem categories. The next part outlines the main factorscausing low rice productivity and cropping intensity in the principal rainfedsubecosystems and the strategies developed for addressing this and other related is-sues. The following section presents selected case study results of the on-farm re-search conducted for developing promising technologies on the basis of environmen-

Planning and managing rice farmingthrough environmental analysisK. Borkakati, V.P. Singh, A.N. Singh, R.K. Singh, A.S.R.A.S. Sastri, and S.K. Mohanty

192 Borkakati et al

tal analysis for rainfed uplands and rainfed lowlands (drought-prone, submergence-prone, and drought- and submergence-prone). The final section draws conclusions onthe planning and management of rice farming in the setting of an agroecologicalframework.

Ecosystem characterization from the mega to micro level in eastern India

An ecosystems analysis for research prioritization at different levels throughout east-ern India has been done collaboratively by the Indian Council for Agricultural Re-search, state agricultural universities, departments of agriculture, and the Interna-tional Rice Research Institute (IRRI 1992).

Mega-level analysisIndia’s rice-growing area occupies 42 million ha. A mega-level analysis indicatedthat, although rice yields in northern and southern India have increased rapidly inrecent years, yields remained basically stagnant in eastern India (except in West Ben-gal, which experienced rapid growth in recent years). The eastern India region, com-prising Assam, Bihar, West Bengal, Orissa, and the eastern parts of Uttar Pradesh andMadhya Pradesh, is the largest rice-growing region in the country and accounts forabout 67% (26.8 million ha) of India’s rice area. In five of the six eastern states,average rice yield (1.8 t ha–1) is below the national average (2.7 t ha–1). About 80% ofrice farming in the region is rainfed. Rainfall is moderate to high, and is limited to ashort period. This results in drought in the uplands and flooding in the lowlands.Sastri and Singh (this volume) give a more detailed hydrological account of easternIndia.

Eastern India is the priority region for research because of its large rice area andlow and stagnant rice yields. It is also a priority region because about half of thecountry’s population of one billion people live here and are largely dependent on ricefarming.

Macro-level analysisThe macro-level analysis of rice-growing ecosystems in eastern India revealed thatonly 21.2% (5.69 million ha) of the 26.8-million-ha rice area is irrigated (IRRI 1992).About 16.4% (4.38 million ha) is upland, 47.8% (12.78 million ha) is rainfed lowland(0–50-cm water depth), and the remaining 14.7% (3.95 million ha) is under thedeepwater (50–100-cm water depth) or very deep water (>100-cm water depth) eco-system category. For the rainfed lowland ecosystem, about 83% (10.6 million ha) hasa shallow water depth (0–30 cm) and 17% (2.2 million ha) has a medium water depth(30–50 cm) during the rice-growing season.

Analysis of drought and flooding patterns, water balance, selected land charac-teristics, and length of the growing season in the shallow rainfed lowland ecosystemshowed that 54.6% (5.7 million ha) is drought-prone, 25.5% (2.7 million ha) is drought-and submergence-prone, 10.3% (1.1 million ha) is submergence-prone, and 9.6% (1.0million ha) is favorable. The entire area in the medium-depth category of the rainfed


lowlands is submergence-prone. There is a wide fluctuation in the extent of thedeepwater ecosystem and the medium-depth category of rainfed lowlands, dependingon rainfall pattern and amount, and onset and cessation of the monsoon.

Rice yields in all the rainfed ecosystems are low and vary greatly from year toyear. Yields in the irrigated areas average 3.2 t ha–1. Yield is 0.6–1.5 t ha–1 in theuplands, 0.9–2.4 t ha–1 in the rainfed lowlands, and 0.9–2.0 t ha–1 in the deepwaterand very deep water areas. The ecosystem with first priority in eastern India is therainfed lowlands because of its area, larger dependent population, and potential foryield increase.

Meso-level analysisA meso-level analysis of rainfed rice ecosystems was conducted in several districts ofeastern India, such as Bahraich (Singh and Pathak 1990, IRRI 1992) and Faizabad(IRRI 1992) of Uttar Pradesh, Hazaribagh (IRRI 1993) of Bihar, and Raipur (IRRI1998, Singh et al 1999) of Madhya Pradesh. Characterization of the rice environ-ments of Faizabad District (total area of 451,100 ha and rice area of about 181,000ha) was done using satellite remotely sensed data, selective field checks, and auxil-iary data (Singh and Singh 1996, Singh 1996). Maps (1:250,000 scale) were preparedto delineate physiographic units, land-use patterns, soils, flooding, and drought. In-formation on climate, groundwater, irrigation sources, landholding, and input usewas integrated with the maps.

The classification of rainfed rice environments showed that about 40% of thearea in Faizabad District (Fig. 1) is favorable rainfed lowland, 51% is drought-prone

Fig. 1. Rainfed lowland rice-growing subecosystems in Faizabad, Uttar Pradesh,India.

Rainfed shallow drought-prone

Rainfed shallow submergence-prone

Rainfed shallow favorable

Pond/lake/river

District headquarters

0 10 20 30 km

194 Borkakati et al

lowland, 2% is submergence-prone lowland, and 4% is submergence- and drought-prone lowland. Apart from drought and submergence, soil sodicity was identified asanother priority research area in the district.

Similarly, a detailed meso-level analysis was done in the Masodha block ofFaizabad District (IRRI 1992). The block covers about 21,000 ha of total land areaand has about 8,000 ha of rice area. Rice-growing environments in terms of physiog-raphy, land use, soils, flooding, drought, groundwater, and irrigation were studied indetail using remote sensing and conventional data. The major part of the block isclassified as a shallow favorable rainfed lowland rice subecosystem. This analysisshowed that about 14% of the block area is affected by flooding, 10% by sodicity, and2% by waterlogging. Only 32% of the groundwater potential has been developed sofar. Recharge-draft analysis showed that about 16,000 ha-m of groundwater is stillavailable for irrigation.

Micro-level analysisWithin each of the upland, rainfed lowland, and deepwater ecosystems in easternIndia, target environments were characterized at the micro level to set research priori-ties within and among the dominant farming systems. More than 100 sites were ana-lyzed.

Rapid rural appraisal techniques, which included agroecosystems mapping anddiagnostic surveys, were employed at all sites (IRRI 1990). The analysis focused onspatial, temporal, resource flow, and decision patterns. The methodology involved atwo-tier training program for researchers on how to set priorities using agroecosystemsanalysis. The problem diagnosis and research prioritization at this level were con-ducted by multidisciplinary teams, with continuous involvement and interactions fromgroups of farmers.

At all sites, the static factors studied were land types, land use, source of watersupply, and soil properties, as described in Singh et al (1993). The dynamic factorswere rainfall and field-water depth; cropping pattern and crop calendars, crop yields,varieties and management practices, insects, diseases, and weeds; production costsand returns; and labor supply pattern, income distribution, landholding size, and de-mography by social class and gender.

The geographic area was zoned into agroecosystems and the problems and op-portunities elucidated in each major agroecosystem (Singh et al 1993). The highestpriority was given to the agroecosystem with the largest rice area. The research pro-grams were then prioritized on the basis of the physical extent (coverage); number ofaffected households; complexity, severity (crop-loss estimates), and frequency of prob-lem occurrence; importance of the affected enterprise in the farming system; and thefarmers’ perceptions of the problem.

All site studies within each subecosystem and ecosystem were pooled and com-pared to identify the commonality of problems and opportunities. This provided anempirical picture of the entire ecosystem, which served as a basis in formulating aneed-based research agenda for developing appropriate technologies for the specific


situations and the allocation of resources at the national, regional, and zonal level(IRRI 1992).

Constraints to rice productivity and strategies for addressing the major issues

The constraints to rice productivity in eastern India vary from state to state and evenfrom area to area. The major constraints to higher rice productivity in different riceecosystems are related to hydrology (moisture stress and flooding), soil and nutrientmanagement, the availability of situation-specific improved varieties and high-qual-ity seed, insect, disease, and weed management, crop establishment, and other spe-cific technologies. These constraints can be listed as follows:

1. Moisture stress due to erratic and often inadequate rainfall, high runoff, poorsoils, and lack of facilities for rainwater and soil moisture conservation/supplementary (life-saving) irrigation (upland and drought-prone rainfedlowlands).

2. Intermittent moisture stress due to low and erratic rainfall; poor soils as inMadhya Pradesh, Orissa, and some parts of Uttar Pradesh; and flash floodsand waterlogging/submergence due to poor drainage, low-lying physiogra-phy, and high rainfall in submergence-prone lowlands, as in Assam, WestBengal, and north Bihar. Accumulation of toxic decomposition products inill-drained soils and soil reduction, encouraging problems of iron toxicity inAssam.

3. Continuous use of traditional varieties because of the nonavailability of seedsand farmers’ lack of awareness about high-yielding varieties (uplands, rainfedlowlands, and deepwater areas).

4. Low soil fertility due to soil erosion, leading to losses of soil nutrients andlow and imbalanced use of fertilizers in uplands, and to the nonavailabilityof a suitable method for applying the fertilizer in standing water in rainfedlowland, semideep, and deepwater areas.

5. Heavy infestation of weeds and insect pests such as blast and brown spotand poor attention to their timely control (uplands and rainfed lowlands).

6. Poor crop stand establishment because of broadcast seeding, resulting inuneven germination (upland and direct-seeded lowlands); and delay in mon-soon onset, often leading to delayed and prolonged transplanting and subop-timum plant population (mostly in rainfed lowlands).

7. Poor adoption of improved crop production technologies because of tech-nology inappropriateness and economic backwardness of the farmers (up-lands and lowlands).

Strategies to increase rice productivity and cropping intensity in eastern Indiamainly included the following:

1. Adoption of runoff rainwater management practices suited to the conditionsof individual farm holdings as well as the watershed as a whole, thus moti-vating farmers to provide life-saving irrigation to the crop during long dryspells.

196 Borkakati et al

2. Emphasis on balanced use of plant nutrients along with the popularization ofintegrated nutrient management approaches and methods of applying requirednutrients in standing water, such as deep placement of urea super granules.

3. Ensuring timely and adequate availability of inputs such as seeds, fertilizers,and credit to farmers. In this regard, multiplication of seeds of promisinghigh-yielding varieties for specific ecologies, such as the rainfed uplandsand drought-prone or submergence-prone lowlands with different water depthand deep water, by the seed-producing agencies and making them availableto farmers can play a significant role in enhancing the productivity of rainfedrice.

4. Promotion of an integrated pest management approach for the control ofinsect pests, diseases, and weeds.

5. In upland rice areas, line sowing may be popularized through suitable seed-ing devices to establish the desired level of plant population, ease in weedcontrol, and the application of other management technologies.

6. Encouraging the use of improved farm implements for effective and timelyfield operations.

7. Organization of field demonstrations of improved technological packages inspecific situations and training of farmers for effective transfer of newlydeveloped crop production technologies. Suitable technological packagesfor different ecosystems by states in eastern Indian are described in Singhand Singh (2000).

Rainfed rice-farming systems technologiesThe main environmental stresses identified in eastern India are drought, submergence,flash flooding, and stagnant deepwater situations. In each of these conditions, tech-nology development aspects considered cropping systems, improved genotypes, wa-ter, nutrients, weeds, and other insect pest management options in an integrated man-ner. In addition, possibilities were also explored for developing water resources,groundwater and rainwater management for drought-prone areas, and drainage op-tions and water recycling in submergence-prone areas. Such strategies were adoptedthroughout eastern India by all the participating centers in upland, rainfed lowland,and deepwater ecosystems.

Technologies for rainfed uplandsIntegrated development of technologies for rainfed rice-farming systems in drought-prone uplands. The agroecological analysis of upland rice areas in Hazaribagh, Bihar,indicated that drought was one of the major causes of low and unstable rice yields andlow cropping intensity. To develop drought management strategies, drought was char-acterized according to duration and severity by analyzing meteorological data from1913 to 1987 from the India Meteorological Department and the Soil ConservationResearch and Demonstration Farm at Hazaribagh. The mean dates of onset and termi-nation of effective monsoon, duration of the monsoon period, and the probability


estimation of drought were determined from the daily rainfall data from these 75years.

The detailed analysis of drought was done for a 25-y period by following theUniversal Hydrologic Equation to arrive at the weekly water balance (rainfall-poten-tial evapotranspiration). Because of the lack of data, however, the analysis did nottake into account the field-stored moisture. The drought analysis was done for theperiod between the 22nd and 43rd wk of the year (28 May to 28 Oct, 154 d duration),as upland rice crop cultivation and the monsoon in this region are confined to thisperiod. During the rice cultivation period, the duration and probability of droughtoccurrence were analyzed separately for three groups of 7 wk each: 22nd-28th wk,29th-35th wk, and 36th-42nd wk, termed as initial, intermediate, and terminal stagesof rice growth. Based on drought occurrence and severity analysis, on-station and on-farm experiments were conducted on adjusting seeding dates to avoid drought in thelatter part of the growing season, selection of short-duration rice cultivars (similar totraditional ones) suitable for drought situations, and weed management and sowingmethods.

Table 1 presents the results of the analyses, which indicate a higher probabilityof drought occurrence at the initial and terminal stages than at the intermediate stageof the crop. Drought affecting the crop at the initial and terminal stages of growth,

Table 1. Selected monsoon and drought characteristics at Hazaribagh,Bihar, India.

Monsoon characteristicsa Amount or time

Normal annual rainfall (mm) 1,299Normal seasonal rainfall (mm) 1,169Probable annual rainfall (mm)

At 99% probability 759At 20% probability 1,385

Probable seasonal rainfall (mm)At 99% probability 666At 20% probability 1,385

Monsoon onset (date) 18 June (earliest 2 June,latest 6 July)

Monsoon termination (date) 11 Oct (earliest 9 Sept,latest 28 Oct)

Monsoon duration (d) 117 (shortest 77, longest 137)Seasonal drought durationb (wk)

Initial stage (28 May to 3.7515 July)

Intermediate stage (16 July 2.00to 17 Sept)

Terminal stage (18 Sept 4.55to 28 Oct)

aPeriod of analysis from 1913 to 1987 using rainfall data. bPeriod of analysisfrom 1972 to 1991 using rainfall and evapotranspiration data.

198 Borkakati et al

however, depended on sowing time and the duration of the varieties used. Almost allfarms in the region experienced drought at the intermediate stage.

Farmers in this region normally do final land preparation and sow rice in the3rd week of June after the onset of effective monsoon. This results in an avoidableloss of initial monsoon moisture to some extent and in delayed sowing in years ofcontinuous rains because of poor workability of the soil. In such cases, crops aremore likely to suffer from drought at the terminal stages of growth. Advance sowingbefore the onset of monsoon was therefore considered as one of the strategies foravoiding or minimizing the effects of terminal drought, which is of longer durationand has a higher probability of occurrence (Table 1). The other strategy in this respectwas to use the selected rice cultivars of short maturity duration (90–100 d) that hadshown consistently better performance in previous on-station and on-farm experi-ments. The results of advance sowing, done on around 7 June before the onset ofrains, indicated significantly higher grain yield, number of panicles m–2, number offertile spikelets panicle–1, and taller plants in all the genotypes tested than sowingdone around 22 June after the onset of rain (normal practice) (Table 2). Among thegenotypes, Brown Gora and Kalinga III were inferior to RR167-982 (Vandana) andRR165-1160 in all respects including panicle weight and panicle length (Table 2).

The early sown plots, however, had a higher weed infestation at the early stagesof plant growth owing to the simultaneous emergence of weeds with rice upon thefirst rain showers. In normal sown plots, the weeds had germinated before rice seed-ing and were destroyed with the additional harrowing. Therefore, weed managementwas followed along with advance sowing in successive experiments that also includedseeding method and rice plant population studies.

With advance sowing, grasses dominated the weed flora in the initial stages ofcrop growth followed by dicotyledons in the latter stages. The bulk of the weed floraconstituted Cyperus rotundus, Cyperus iria, Echinochloa colona, Cynodon dactylon,Setaria glauca, Commelina benghalensis, Aeschynomene indica, and Brachiariaramosa, which competed with rice during all its growth stages. The population ofAgeratum conyzoides was severe at the reproductive stage only. Losses in rice grainyield (difference between a weed-free and weedy plot) were as high as 77%, and werehigher in drilled than broadcast-seeded crops. However, when manually weeded at 20and 40 days after sowing, the drilled crop produced significantly higher yield (2.4 tha–1) than the manually weeded broadcast crop (2.0 t ha–1). Application of butachlorwith one handweeding was as effective as two handweedings in both systems.

The initial 4-wk period of crop growth was crucial for weeding in the case ofearly maturing tall genotypes, such as Kalinga III, having poor early vigor in contrastwith the semitall genotypes, such as RR167-982 (Vandana) and Brown Gora, thathave higher early vigor and a suppressive effect on weeds.

Plant population, attained using a seeding rate of 500 seeds m–2 or 112.5 kgha–1, and closer row spacing (20 cm) also had a suppressive effect on weeds, as re-flected by lower weight of total weed dry matter. Rice grain and straw yields werehigher under these practices than using a higher or lower seeding rate (400 or 600seeds m–2) and sowing by broadcast or with wider row spacing (30 cm).


Table 2. Effect of date of sowing, weed management, seed rate, and sowing method on thegrowth and yield of upland rice, Hazaribagh, Bihar, India.

Treatment Grain yield Plant height Panicles Fertile spikelets(t ha–1) (cm) (no. m–2) (no. panicle–1)

Date of sowing (variety RR167-982)a

7 June 2.2 87.1 325 72.422 June 1.8 77.9 309 65.4

Cultivarsb Grain Plant Panicle Panicle Filled 1,000-grain(sown on 7 June) yield height length weight grains (no. weight (g)

(t ha–1) (cm) (cm) (g) panicle–1)

Brown Gora (farmer’s var.) 1.2 111 16.4 1.49 41 28.5Kalinga III 1.5 106 18.9 1.50 55 22.7RR167-982 (Vandana) 2.5 117 19.5 2.17 76 24.2RR165-1160 2.1 106 23.8 2.50 97 21.3

Weed managementc (variety RR167-98, sown on 7 June) Grain yield (t ha–1)

Farmer’s method in broadcast 1.02Weedy check broadcast 0.75Weedy check drilled 0.54Weed-free broadcast (handweeded) 1.97Weed-free drilled (handweeded) 2.37Butachlor + 1 handweeding (30 d after sowing) 2.01

Seed rateb Grain yield Sowing method Grain yield(t ha–1) (variety RR167-982, early sown on 7 June) (t ha–1)

400 seeds m–2 2.00 Broadcast 1.8(90 kg ha–1)

500 seeds m–2 2.25 Sowing behind plow 2.2(112.4 kg ha–1) (20-cm rows)

600 seeds m–2 2.10 Sowing behind plow 1.7(135 kg ha–1) (30-cm rows)

aMean of three kharif (wet) seasons, 1989-91. bMean of two kharif seasons, 1991-92. cMean of three kharifseasons, 1990-92.

Sowing in furrows at 20-cm spacing behind the plow using a rate of 112.5 kgseed ha–1 was superior to broadcast not only for ease in weed control but also forcombating drought effects, especially at the initial stages. Furrow sowing allowedseeds to be placed at 3–5-cm depth, thus encouraging better root development andexploitation of soil moisture. Soil moisture retention and release characteristics alsosupport this and indicate that the upper 5-cm layer of these soils dried up quickly evenwhen their subsoil layers had adequate moisture. Sowing in furrows also protectedseeds from desiccation and bird damage and required a lesser amount of seed thanthat used by farmers in broadcasting (150–200 kg ha–1).

200 Borkakati et al

Several farmers in the on-farm study villages have shifted to furrow sowingfrom broadcast and have improvised the “country plow” by removing its iron shear toopen only 5–6-cm-deep furrows. They have also replaced traditional cultivars withRR167-982 (Vandana) and have opted to advance seeding to before the onset of mon-soon.

Managing rainwater for stable rice yields and improved cropping intensity.Rainfed uplands in eastern India cover several million ha and are contiguous in theChotanagpur plateau region. Because of low and concentrated rainfall (1,130 mmfrom June to October), these lands are traditionally cultivated only in the wet seasonto grow a crop of millet-blackgram-rice-fallow in a 4-y crop rotation cycle. Theseareas generally remain fallow during the dry season because of the lack of soil mois-ture and unavailability of irrigation water.

Analysis of the existing farming systems and hydrological mapping of the areaindicated higher potentials for rainwater collection in valley check reservoirs that canbe used to supplement the water requirement of rice during drought spells and forirrigating short-duration dry-season crops. The farming systems analysis includedland, labor, and cash resources; land types and their use pattern; and crop calendarswith respect to rainfall pattern and labor availability and use. Hydrological mappingincluded studies on available water resources, surface and subsurface hydrology, waterlosses and recharge characteristics of the soil, occurrence and severity of drought,and climatic factors (Paul and Tiwari 1994).

The possibility of on-farm rainwater collection and management was exploredat Handio village in Hazaribagh. Starting with two farmers as pilot cases, villagersconstructed 16 rainwater-holding structures (valley check reservoirs and dug ponds)ranging from 1.75 ha in the uppermost toposequence with a large runoff catchmentarea to 0.25 ha in the lower lands along the slope (Singh et al 1993). The constructionof tanks was done through family and communal labor on an exchange basis duringthe summer months as sufficient labor was available in this period. The catchment tostorage ratio in terms of land surface area varied from 10:1 to 6:1 depending on thegeneral topography and micro relief of the area. On average, the volume of earth dugand moved amounted to 0.6 to 1.0 m3 capita–1 d–1. The dug soil was placed on thedikes and compacted by the respective farm families themselves. The volume of avail-able water in these structures varied from 9,100 m3 to 16,900 m3 for 7 to 10 mo (Julyto April). In addition to the rainwater-holding structures, the farmers have also dugand constructed 37 cement wells in their respective landholdings with governmentsupport.

The water from tanks is distributed for irrigation by gravity through cementculverts placed at different depths from the water surface and connected to earthenchannels at different elevations. This allows farmers to draw water at different levelsin the tank without cutting the dike or siphoning. It also reduces maintenance of dikes.

With these practices, cropping intensity in the village has increased from <100%to 160% in the upper toposequence and from 150% to almost 250% in the lower lands(IRRI 1992). The increase in cropping intensity was achieved through vegetable cul-tivation during the dry season that includes table pea-potato + coriander and chillies


from October to February and a summer crop of tomato for some farmers (Singh et al1993). All farmers continue to grow rice, blackgram/pigeonpea, and some vegetablesduring the wet season with stable yields. Of the various crop sequences, rice-table peaor niger-potato + coriander-tomato are the most common in the village.

With the availability of water, some farmers have started growing fish (com-mon carp) at a low stocking density (6,000 ha–1). Results with older fingerlings ofheavier weight (50 g piece–1 or more) are reported to be more promising. With theincrease in cropping intensity, farmers in the village have also started using animalmanure in vegetable production and stall feeding of livestock, a practice not followedearlier because of difficulties in dung collection from open land grazing at distantplaces.

Managing groundwater resources to alleviate drought effects. Drought is a majorcause of low and unstable yields and low cropping intensity in Hazaribagh, Bihar. Toalleviate drought effects, farmers in this area adopt various cropping and agronomicpractices and explore ways to provide supplemental irrigation. They dig shallow wellsand construct rainwater-holding structures (on-farm reservoirs and tanks). The water-supplying capacity of the wells, however, is highly variable in the district, dependingon the local recharge capacity and the draw of water. Some wells remain productivethroughout the year, whereas others dry up after a few months. It is also not possibleto have productive wells in all places.

The State Remote Sensing Application Center (IRRI 1993) prepared a ground-water potential map for Hazaribagh covering a geographical area of 1,116,500 ha anddivided the district into four groundwater zones: poor, poor to moderate, moderate,and moderate to good. The map provided only a qualitative assessment and could notbe used directly for further groundwater exploitation purposes because no informa-tion on cropped area was available on the map. The local unit of the Central Ground-water Board had calculated the net groundwater recharge and draft in the district,taking the block as a unit, but it did not provide spatial information on the promisingareas for groundwater exploitation within the block. A study was therefore carried outusing satellite remote sensing to generate information on rice and other crop landsand integrate it with the available groundwater information and block and villageboundary maps on a geographic information system (GIS) to identify the promisingcropland areas for further groundwater exploitation.

A crop area map of the district was prepared based on the interpretation ofIndian Remote Sensing Satellite (IRS-IA) images of three different years (1988, 1991,and 1992). Cropped area analysis showed that approximately 48% of the district areais under agriculture and about 49% is under forest. Nearly all agricultural land (532,550ha) is cropped during the kharif (rainy) season. About 25,620 ha of land are croppedduring the rabi (dry) season. Rice is the dominant kharif crop. Cropping intensity inthe district varies from 105% to 116%.

Integration of the crop-land map, groundwater zone map, and block boundarymap resulted in a map showing the distribution of groundwater potential in crop landsof different blocks. Moderate and moderate-to-good groundwater potential zones weremerged, since these were considered promising from the viewpoint of groundwater

202 Borkakati et al

Hunterganj

Partabpur

Simaria

Chatra Itkhori

Chauparan

Domchanch

Sargawan

Markacho

Ichak

BishungarhHazaribagh

Tandwa

Keredari

Barkagan

Churchu

Crop lands with moderate to good groundwater potentialBlock boundary

Barhi

Mandu

GolaPatratu

Ramgarh

exploitation. Poor and poor-to-moderate groundwater zones were excluded due totheir low potential. The final map (Fig. 2) shows promising areas for groundwater usein crop lands of different blocks. The extent of these areas was then calculated (Table3). The calculations showed that, out of 24 blocks of the district, 11—Chatra, Barhi,Churchu, Patratu, Ramgarh, Gola, Mandu, Barkatha, Jainagar, Markacho, andKodarma—had more than half of their crop-land area as promising for groundwaterexploitation. The remaining 13 blocks showed less than half of their crop lands aspromising.

The data on annual groundwater recharge, draft, and available balance in dif-ferent blocks (Table 3) showed that the annual recharge varies from 8.5 million m3 inJainagar to 51.0 million m3 in Simaria. The net groundwater draft is low, rangingfrom 0.8 million m3 in Churchu to 7.1 million m3 in Hunterganj. The draft is high,however, in Hazaribagh (12.1 million m3). Since the geographical area of differentblocks varied considerably, from 228 km2 in Jainagar to 976 km2 in Simaria, theavailable groundwater balance was averaged by calculating the balance in each km2-area of the block. Overlaying this information on the promising groundwater area

Fig. 2. Promising area for groundwater exploitation in crop lands under differentblocks of Hazaribagh District.


map showed that 10 of the 11 promising blocks (except Ramgarh) had a net balanceof more than 30,000 m3 km–2 and are suitable for further groundwater exploitation incrop-land areas as identified. Overlaying village boundaries on the block map led tothe identification of promising villages for groundwater exploitation. Possibilities fordeveloping surface irrigation sources were indicated in the remaining 14 blocks ofthe district.

Technologies for the drought-prone rainfed lowland ecosystemAt Raipur, which represents the drought-prone lowland ecosystem, the duration ofhumid (rainfall (R) > potential evapotranspiration (PE)) and moist (PE>R>PE/2) pe-riods is 151 d (Sastri and Singh, this volume). This suggest that, if an early or me-dium-duration (110–125 d) rice crop is grown, especially in heavy soils, a secondcrop of chickpea or lathyrus or linseed can be grown with conserved moisture. Thesecrops, unlike rice, can also thrive even in a submoist (PE/2>R>PE/4) period.

Table 3. Promising areas for groundwater exploitation in different blocks ofHazaribagh District, Bihar, India (1992-93).

% Net annualTotal crop land recharge Net draft Groundwater

Block crop land with available (million m3) balance(ha) promising (million (m3 km–2)

groundwater m3)

Pratappur 18,073 13 28.5 1.2 40,700Hunterganj 21,171 1 28.8 7.1 42.300Chatra 21,751 58 36.4 2.5 53,100Simaria 26,432 43 51.0 2.3 50,200Tandwa 19,659 13 19.6 1.3 48,800Keredari 14,916 26 20.1 1.0 44,000Barkagan 18,385 19 17.1 1.5 34,900Katkamsandi 21,107 26 17.7 3.5 30,600Itkhori 25,317 16 27.1 2.8 45,600Chauparan 24,433 40 33.8 4.2 44,333Barhi 20,603 98 27.2 1.9 52,300Ichak 17,038 10 17.2 1.6 39,600Hazaribagh 15,961 2 17.3 12.1 16,400Churchu 17,716 74 20.2 0.8 46,500Patratu 16,412 82 13.0 2.0 35,100Ramgarh 22,396 97 11.7 5.2 21,300Gola 25,580 80 18.5 2.6 47,300Mandu 22,634 96 28.6 3.9 57,200Bishungarh 25,524 2 25.0 2.8 42,700Barkatha 20,432 97 14.2 1.3 30,000Jainagar 19,391 100 8.5 1.8 29,400Markacho 17,828 95 14.7 1.0 43,200Kodarma 18,546 74 19.6 4.0 33,900Satgawan 6,873 nil 13.5 1.6 39,300

204 Borkakati et al

With this hypothesis, on-farm experiments were conducted in nine farmers’fields in black soils from 1995-96 to 1998-99 (4 years) in Tarpongi village about 30km north of Raipur. Second crops of chickpea, lathyrus, and linseed were grown afterearly (var. Poornima) and medium-duration (var. Mahamaya) rice varieties. Tables 4and 5 show the results of the four experiments, which indicate that

Table 4. Average data of the double-cropping experiment.

Crop sequence Mean grain yield (t ha–1)

Wet Winter Wet- Winter- Total valuea ofseason season season season produce (Rs ha–1)

crop crop

Early rice Chickpea 2.3 0.25 16,200(Poornima) Lathyrus 2.3 0.12 14,400

Linseed 2.3 0.05 14,460

Medium rice Chickpea 3.2 0.13 20,760(Mahamaya) Lathyrus 3.2 0.07 19,980

Linseed 3.2 0.02 19,800

aPrices for rice: Rs6,000 t–1; lathyrus: Rs8,000 t–1; chickpea: Rs10,000 t–1;linseed: Rs15,000 t–1.

Table 5. Economics of rice-chickpea crop sequence at Raipur, MadhyaPradesh.

Mean yield (t ha–1) GrossYear Crop sequence incomeb

Rice Chickpea (Rs ha–1)

1995-96 Early rice-chickpea 3.1 0.42 23,040Medium rice-chickpea 4.5 0.25 29,616

1996-97 Early rice-chickpea 1.6 0.00a 9,720Medium rice-chickpea 2.5 0.00a 14,820

1997-98 Early rice-chickpea 2.3 0.27 16,500Medium rice-chickpea 3.4 0.06 20,940

1998-99 Early rice-gram 2.1 0.33 15,660Medium rice-gram 2.5 0.22 17,760

Mean Early rice-gram 2.3 0.26 16,230Medium rice-gram 3.2 0.13 20,784

aCrop could not be established because of inadequacy of moisture. bPrices forrice Rs6,000 t–1 and chickpea Rs10,000 t–1.


1. There are possibilities of growing a second crop after rice with the con-served moisture and nutrients.

2. Though the productivity of a second crop (chickpea or linseed or lathyrus) ishigher in short-duration rice fields than in medium-duration rice fields, theoverall rice equivalent productivity of the crop sequence (total cash value ofproduce from both the crops, based on prevailing price) is higher in the caseof medium-duration rice followed by a second crop, indicating that this sys-tem is more suitable.

3. The rice equivalent productivity (cash value) of chickpea produce is higherthan that of either lathyrus or linseed.

Table 5 indicates that in general chickpea yield was higher in the short-durationrice-chickpea crop sequence, but the average total income in this sequence is Rs16,230,whereas, in the medium-duration rice-chickpea crop sequence, the average total in-come is Rs20,784. It is therefore recommended that, in view of the higher productionpotential, the medium-duration rice-chickpea system should be followed in heavy(black) soils of the drought-prone lowland rice ecosystem. Year-to-year fluctuationsin the productivity of both crops occur, however. During 1996-97, which happened tobe a severe drought year, the chickpea crop could not be sown due to dry soil condi-tions and the rice yields were the lowest of the four years. The period 1995-96 was agood rainfall year and both the chickpea and rice yields were the highest of the fouryears. A risk analysis on the cropping systems over time is essential in such cases.

In drought-prone lowland ecosystems, the ill effects of drought can be allevi-ated with timely and proper crop management practices. In the on-farm experimentsconducted at Raipur during 1995-96 to 1998-99, the highest, lowest, and median riceproductivity across fields and its standard deviation are shown in Table 6.

Table 6 shows that, even in bad (drought) years such as 1996, the highest yieldsof rice varieties Mahamaya and Poornima were 3.6 and 2.7 t ha–1, respectively, whereastheir lowest yields during 1996 were 1.8 and 0.4 t ha–1, respectively. Such a variationin yields across fields is a clear indication of crop management differences, which inthis case were mainly related to the timeliness of crop establishment, “beushening”(dry-sown crop then plowed and laddered 30–35 d after sowing with 10–15 cm ofstanding water), and weed control operations. The beneficial effects of proper man-agement can also be realized even in good rainfall years such as 1995 (Table 6).

An analysis of the dry weeks (weeks receiving less than 50 mm rainfall) indi-cates that 1999 had a continuous dry spell of 10 wk from 27 August to November, thatis, from the 35th to the 44th standard meteorological week (SMW). This resulted insevere drought conditions. In light soils, the rice crop failed completely. In heavysoils, some farmers provided a small supplementary life-saving irrigation from thewater collected in roadside ditches, which benefited those farmers considerably (Table7).

Technologies for the submergence-prone rainfed lowland ecosystemThe agroecological analysis of rainfed lowland rice areas in Jorhat District of Assamindicated that submergence proneness was one of the major causes of low and un-

206 Borkakati et al

stable yields of rice. Farmers in these situations normally practice rice monocropping,mainly in the sali (winter) season. The two-rice cropping pattern is also followed insome villages, but to a limited extent. For the sali crop, seedbed preparation starts inJune with the onset of monsoon and transplanting is done in July. But, because offloods in June, transplanting is delayed, which, on the recession of floodwater, con-tinues up to August. Because of late planting, however, the flowering stage coincideswith low temperature. As a result, grain filling is not completed, which leads to lowerrice yields. Also in these situations, farmers use very low rates of fertilizer (15-5-0 N,P2O5, K2O ha–1, respectively).

Table 7. Rice grain yield (t ha–1) of a short- andmedium-duration rice variety under severedrought conditions with and without supplemen-tary irrigation (SI) in a heavy soila, Raipur, India,1996.

Rice variety Grain yield (t ha–1)

Medium duration– with SI 2.57– without SI 2.02

Mean 2.48

Early duration– with SI 1.87– without SI 0.41

Mean 1.62

aThere was no grain yield in light soils for any of the ricevarieties under any of the treatments.

Table 6. Rice productivity variations using two varieties at Raipur,Madhya Pradesh.

Rice Productivity (t ha–1) StandardYear variety deviation

Highest Lowest Median (t ha–1)

1995 Mahamaya 5.8 3.3 4.5 1.1Poornima 3.7 2.5 3.1 0.5





To develop management strategies, a detailed analysis of climatic water bal-ance was done for 18 years (1980-97). The depth of surface water in 80 well-distrib-uted paddy fields in four selected villages (20 fields in each village) representing thedistrict was measured twice a week from 2 June to 27 November (178 d duration) in1997, as sali rice crop cultivation and the monsoon in this region are confined to thisperiod.

Sastri and Singh (this volume) give details of an agroclimatic inventory in Assam.Figure 3 shows the climatic water balance of bunded rice fields using average rainfalland PE values as inputs. This was computed using the bookkeeping procedure ofThornthwaite and Mather (1955). The following assumptions were made:

1. The water-holding capacity of the soils up to saturation is 300 mm.2. From the surplus water that accumulates in the rice fields, percolation losses

occur at the rate of 4 mm d–1, or 120 mm mo–1.3. After percolation, the excess water is stored in the fields as standing water

up to 50-mm depth in rice fields.4. Any water depth above 50 mm is considered as floodwater.The soil moisture recharge in Jorhat, Assam, starts on 18 March and continues

till mid-June (Fig. 3). From mid-June onward, percolation losses continue and stand-ing water up to 50-mm depth remains till 20 October. The flood period begins atJorhat after 15 June and ends by mid-August; its peak under normal conditions isaround mid-July. The average amount of floodwater in rice fields (above standingwater of 50 mm) is 112 mm.

In all the villages, the water started accumulating from 4 June (24th SMW). Itsdepth fluctuated after 4 June and peaked on 13 August (33rd SMW) (Table 8). Then itdeclined and reached a minimum in October. The surface-water depth was directlyrelated to rainfall, which, during the peak water-depth period, was the maximum (307.6mm) of the season. Hydrology varied widely within the villages. Based on the sur-

Fig. 3. Climate water balance (A) and distributions of surplus water (B) at Jorhat, Assam, India.

450

350

250

150

500

J F M A M J J A S O N DMonth

mm

140

120

100

80

60

40

20

0Jun Jul Aug Sep Oct

Month

mm

Percolation

Standing water

Floodwater

Soil moisture useSoil moisture accumulationSurplus waterRainfallPE

A B

208 Borkakati et al

face-water depth, rice subecosystems were delineated in each village. Mainly, threedifferent hydrological situations were identified from the data with maximum waterdepth of 0–10, 10–30, and 30–50 cm during the crop season. The rainfed shallow,submergence-prone (10–30 cm) situation occupied more area than the other two hy-drological (water depth) situations.

Crop production strategies for such cases were adjusted for planting time, se-lection of rice varieties to accommodate two rice crops, and selection of rice varietiesand fertilizer management according to the water-depth classes even withinsubecosystems.

Increasing rice-cropping intensity. To increase cropping intensity, on-farm tri-als were conducted to accommodate two rice crops in the cropping pattern by select-ing relatively higher-yielding and short-duration (early maturing) promising ahu (au-tumn) rice varieties for preflood cultivation. In this effort, some local varieties col-lected from different localities and some high-yielding varieties were evaluated againstthe popularly grown rice variety Rangadoria as the local check. Among the varietiestested, only Culture-1 matured earlier (88 d vs 96 d for the local check) and hadaround a 20% yield advantage over the local check variety (Table 9). Within the samematurity group (about 90 d), varieties such as Chilarain, Luit, and Lachit, developedby Assam Agricultural University, have been found to be very promising. Their aver-age yield surpassed 4 t ha–1. They are now becoming popular among the farmers.

Using these promising varieties, ahu (autumn) rice can generally be broadcast/transplanted in March-April and harvested in June-July before floodwater enters the

Table 8. Surface-water deptha (in cm) in paddy fields in different hydrological situations, Assam,India, 1997.

Hydrological situation classStandard (cm)

meteorologicalweek 0–10 10–30 30–50

24 1.2 2.8 10.825 1.2 2.8 17.226 1.8 3.4 18.227 2.1 4.2 19.628 2.2 4.4 20.229 2.0 4.0 21.430 2.3 4.5 22.531 4.5 6.8 28.532 8.8 12.5 40.633 8.6 14.2 43.534 6.5 11.3 40.235 6.2 12.2 40.836 7.4 9.6 42.2

aValues are means of 46 measurements (twice a week) from a total of 80 paddy fields from four villages (20 ineach village).

Hydrological situation classStandard (cm)

meteorologicalweek 0–10 10–30 30–50

37 6.2 8.5 34.338 4.3 6.2 31.839 2.3 4.8 24.640 0.0 2.1 22.541 0.0 2.0 15.242 0.0 2.5 15.843 0.0 1.8 12.844 0.0 1.0 8.245 1.2 0.0 6.246 0.0 0.0 5.847 0.0 0.0 2.048 0.0 0.0 0.049 0.0 0.0 0.0


rice fields. After harvesting of ahu rice, sali (winter) rice can easily be accommodatedfrom July-August and harvested in November-December.

Management of inputs in different subecosystems (hydrological classes). Ex-periments were conducted with a semidwarf variety (Ranjit) and a tall variety (IET-10016) with farmers’ usual fertilizer dose (15-5-0 kg N, P2O5, K2O ha–1, respectively),and the recommended fertilizer dose (40-20-10 kg N, P2O5, K2O ha–1, respectively)under different hydrological conditions (0–10, 10–30, and 30–50-cm water depth).The results revealed that, in general, variety Ranjit was superior to IET-10016, pro-ducing about 8% higher yield. Application of the recommended level of fertilizershowed a definite yield advantage of about 39% over the farmers’ usual practice in allthe hydrological conditions (Table 10). However, it was more in the case of semi-dwarf variety Ranjit than tall variety IET-10016. This clearly indicates that the inad-equate application of fertilizer is one of the major causes of lower rice productivity.

Depth of surface water also exerted a significant influence on grain yield of riceunder both levels of fertilizer application and for both varieties (Table 10). Althoughan increase in surface-water depth from 0–10 to 30–50 cm had a positive effect on thegrain yield of IET-10016, the same increase in water depth had a negative effect onthe grain yield of Ranjit. For both varieties, however, such hydrological effects weremore pronounced at water depths exceeding 30 cm, particularly at the recommendedlevels of fertilizer application. Up to 30-cm water depth, there was a slight change inthe grain yield of both varieties, but at water depths exceeding 30 cm the change ingrain yield was substantial.

Although both varieties responded to the recommended level of fertilizer appli-cation, they behaved differently in surface hydrology class (Fig. 4). Ranjit yieldedhighest in the 0–10-cm water-depth class, whereas IET-110016 yielded the highest inthe 30–50-cm depth class. Both yielded similarly in the 10–30-cm water-depth classwith the farmers’ fertilizer level. These results indicate that, to raise the yields ofrainfed lowland rice, fertilizer-responsive, high-yielding rice varieties are needed ac-cording to the specific hydrological situation (surface-water-depth class).

Table 9. Yielda of preflood ahu (autumn) rice (t ha–1) in the flood-prone ecosystem, Assam, 1998.

Variety Days to maturity Grain typeb Grain yield(t ha–1)

Gunilahi 100 SB 1.7Megli 106 SB 2.4Isahajay 100 SB 2.1Culture-1 88 LB 2.3Pusa 2-21 115 LB 2.3Annada 115 LB 2.7Rangadoria (local check) 96 SB 1.9

aMean of 8 replicates. bSB = short bold, LB = long bold.

210 Borkakati et al

Selection of suitable rice varieties. The results of the environmental analysishave been used to help farmers select rice varieties according to ecosystem suitabilityin 20 villages of Milkipur block in Faizabad District, Uttar Pradesh. A micro-levelrice ecosystem analysis was first carried out in these villages. Aerial photographswere used to delineate landforms, land use, and hydrological conditions in these vil-lages. Within the culturable agricultural lands (including presently uncultivated butpotentially cultivable), the percentage of area under different land types was uplands12%, midlands 78%, and lowlands 10%. The meso-level (district-level) analysis ofthe area had shown these villages to be under the rainfed shallow (0–30-cm water

Table 10. Grain yielda (t ha–1) of rice varieties as affected by fertilizer manage-ment at different water depths, Assam, 1998.

Variety (V)Factor

Ranjit IET-10016 Mean

Fertilizer level (F) in kg N, P2O5, K2O ha–1

Farmers’ fertilizer levelb 3.2 3.2 3.2Recommended fertilizer levelc 4.8 4.2 4.5LSD (0.05) F = 0.262 V × F = 0.371

Surface water depth (SWD) in cm0–10 cm 4.7 2.9 3.810–30 cm 4.3 3.5 3.930–50 cm 3.0 4.7 3.8LSD (0.05) SWD = 0.457 V × SWD = 0.647

Fertilizer × water depth × variety0–10 cm: Farmers’ levelb 3.7 2.7

Recommended levelc 5.8 3.1

10–30 cm: Farmers’ levelb 3.5 3.1Recommended levelc 5.1 3.9

30–50 cm: Farmers’ levelb 2.4 3.9Recommended levelc 3.5 5.5

LSD (0.05) V × F × SWD = 0.643Mean of varieties 4.0 3.7LSD (0.05) V = 0.335

Fertilizer level on surface-water depth Farmers’ fertilizer Recommendedlevel level

0–10 cm 3.2 4.410–30 cm 3.3 4.530–50 cm 3.1 4.5LSD (0.05) 0.456

aMean of 8 replicates. b15, 5, 0 kg N, P2O5, K2O ha–1. c40, 20, 10 kg N, P2O5, K2O ha–1. dAtrecommended level of fertilizer.


depth) drought-prone lowland subecosystem. The micro-level analysis showed about88% of the rice-growing area under rainfed shallow drought-prone lowlands and theremaining 12% under the rainfed medium-deep (30–50-cm water depth) waterloggedsubecosystem (Table 11). Subecosystem characterization also included informationon drought and flooding pattern, timing, and duration in the rice-growing season.

Based on this knowledge of rice environments, the farmers were provided witha rice variety (NDR-118) that is an early (90 d flowering), fine-grained variety, ex-perimentally found to be very suitable for rainfed shallow drought-prone lowlands.The performance of this variety was compared with that of the farmers’ grown variet-ies prevalent in the area (IR36, Sarju-52, and Pant-4), which are of medium duration(100 d flowering) and generally suitable for irrigated or rainfed shallow favorablelowlands. Mahsuri, a variety of medium to long duration (110–120 d flowering) andtall plant type, was being grown by farmers on medium-deep waterlogged lowlands.Seeds of these varieties were also provided to the farmers. During the wet season of1994, despite a normal total rainfall in the area, terminal drought in rice occurredbecause of no rain after 20 September. Because of this drought, some farmers couldnot get any yield and the crop failed completely. Data collected from 385 farmersfrom these villages showed such total failure to the extent of only 7% of farmers’fields with NDR-118. The other varieties used by farmers showed total failure to theextent of 39% for farmers using IR36, 18% for Mahsuri, and 11% for Sarju-52 (Table12). Data collected from two selected villages in subsequent years showed a signifi-cant increase in rice area coverage under NDR-118, the variety found suitable for thedominant subecosystem of the area.

Fig. 4. Grain yield (t ha–1) of rice genotypes Ranjit and IET-10016 inthree surface-water-depth classes with farmers’ level (15, 5, 0) andrecommended level (40, 20, 20) of N, P2O5, K2O (kg ha–1) applica-tion.

6

5

4

3

2

1

030–50 cm10–20 cm0–10 cm

Water-depth class


Farmers’ level (Ranjit)

Recommended level (Ranjit)

Farmers’ level (IET-10016)

Recommended level (IET-10016)

212 Borkakati et al

This shows that the knowledge of environmental characterization helps in se-lecting appropriate varieties, whose use can help farmers increase production andreduce risk in uncertain rainfed environments.

Conclusions

The results of the studies carried out in eastern India indicate that the performance ofthe rice crop is strongly influenced by the hydrological conditions prevailing in thegrowing season. These hydrological influences can be mitigated considerably if cropmanagement strategies are developed on the basis of the knowledge of environmentalconditions. Likewise, other appropriate technologies can be developed using suchknowledge and can easily be applied by farmers in rainfed situations.

Table 11. Area under different rice ecosystems in Milkipur studyarea, Faizabad District, India.

Rice subecosystems and other Area % geographicalland-use class (ha) area

Rainfed shallow drought-prone lowlands 4,447 62.7Rainfed medium-deep waterlogged

lowlands 583 8.2Other crops 736 10.4Forests and orchards 472 6.7Uncultivated 502 7.1Ponds/lakes 193 2.7Habitation 155 2.2

Total 7,088 100.0

Table 12. Rice varieties tested, ecosystem suitability, and percent-age total crop failure in farmers’ fields in Milkipur area, FaizabadDistrict, India (kharif 1994). Total number of farmers = 385.

Duration and ecosystem % farmersRice variety suitability reporting total

crop failure

IR36 130 d, irrigated/rainfed shallowfavorable lowlands 39

Sarju-52 140 d, irrigated/rainfed shallowfavorable lowlands 11

Mahsuri 145 d, irrigated/rainfed shallowfavorable lowlands/medium-deep waterlogged 18

NDR-118 90 d, rainfed shallow drought-pronelowlands/rainfed uplands 7


ReferencesIRRI (International Rice Research Institute). 1990. Program report for 1989. Los Baños, La-

guna (Philippines): IRRI. 302 p.IRRI (International Rice Research Institute). 1992. Program report for 1991. Los Baños, La-



guna (Philippines): IRRI. 175 p.Paul DK, Tiwarim KN. 1994. Rainwater storage systems for rainfed rice lands of eastern India:

results from research in Hazaribagh district. In: Bhuiyan S, editor. On-farm reservoirsystems for rainfed rice lands. Manila (Philippines): International Rice Research Insti-tute. 164 p.

Singh VP. 1996. Agroecological analysis for sustainable development of rainfed environmentsin India. J. Indian Soc. Soil Sci. 44(4):601-615.

Singh VP, Pathak MD. 1990. Rice growing environments in Baharaich district of Uttar Pradesh,India. Tech. Bull. No. 1. Uttar Pradesh Council of Agricultural Research, Lucknow,India. 56 p.

Singh VP, Singh AN. 1996. A remote sensing and GIS-based methodology for the delineationand characterization of rainfed rice environments. Int. J. Remote Sensing 17(7):1377-1390.

Singh VP, Singh RK, editors. 2000. Rainfed rice: a sourcebook of best practices and strategiesin Eastern India. Makati City (Philippines): International Rice Research Institute.292 p.

Singh VP, Singh RK, Singh RK, Chauhan VS. 1993. Developing integrated crop-livestock-fishfarming systems for rainfed uplands in Eastern India. J. Asian Farm. Syst. Assoc. 1:523-536.

Singh VP, Singh RK, Sastri ASRAS, Baghel SS, Chaudhry JL. 1999. Rice growing environ-ments of Eastern India: an agroclimatic analysis. IGAU and IRRI Pub. 76 p.

Thornthwaite LW, Mather JR. 1955. The water balance. In: Climatology 11:1-14. Drexel Insti-tute of Technology, Centerton, N.J. (USA).

NotesAuthors’ addresses: K. Borkakati, Department of Soil Science, Assam Agricultural University,

Jorhat, Assam, India; V.P. Singh, International Rice Research Institute, Los Baños, Phil-ippines; A.N. Singh, Uttar Pradesh Remote Sensing Applications Center, Lucknow, In-dia; R.K. Singh, Central Rainfed Upland Rice Research Station, Hazaribagh, India;A.S.R.A.S. Sastri, Indira Gandhi Agricultural University, Raipur, Madhya Pradesh, India;S.K. Mohanty, Central Rice Research Institute, Cuttack, Orissa, India.


Agroclimatic inventory for environmental characterization . . . 215

Agroclimatic inventoryfor environmental characterizationof rainfed rice-based cropping systemsof eastern IndiaA.S.R.A.S. Sastri and V.P. Singh

Climate is an important component of environment. In environmental charac-terization for developing strategies to improve productivity, an agroclimaticinventory is a prerequisite. In eastern India, comprising the states of Orissa,Bihar, West Bengal, eastern Madhya Pradesh, eastern Uttar Pradesh, andthe northeastern states, rice is grown mostly under rainfed conditions inupland, lowland, and flood-prone ecosystems. As this region comes underthe influence of the southwest monsoon from June to October, crop produc-tivity depends entirely upon the vagaries of monsoon. Also, because of mon-soon activity, the sky is mostly overcast and radiation becomes a limitingfactor.

In this study, the moisture regime was analyzed using simple measuressuch as amount of rainfall and number of rainy days and derived parameterssuch as moisture availability and stable rainfall periods at different probabil-ity levels. We found that the average rainfall in eastern India matches mostlywith 30% or, in some cases, 40% probability levels, indicating that any aver-age rainfall-based strategy for improving rice productivity is successful onlyonce in three years. Therefore, a concept of stable rainfall period has beendeveloped by defining it as a period when weekly rainfall exceeds 50 mm.Stable rainfall periods even at a 60% probability level are very short in someparts of the region, suggesting a need to develop viable location-specificwater management practices. The thermal regime in this region is not gener-ally a limiting factor, except on a few occasions. Because radiation is a limit-ing factor during the active monsoon months, however, we need to identifyrice varieties with higher energy-use efficiency.

Climate and its components such as temperature and rainfall are the factors that mostaffect agricultural productivity and at the same time are the least modifiable (Frankel1976). Rainfall is the major climatic element that affects crop growth and develop-ment, particularly where rainfed farming is widely accepted. Also, in comparisonwith other climatic factors, data on rainfall are generally available for an extensive

216 Sastri and Singh

network (Sastry 1976). Temperature conditions in a particular geographic area repre-sent decision criteria for the natural establishment of plants. All the physiologicalprocesses take place within certain tolerance ranges of temperature. The temperaturehas to reach a certain minimum value at which biological processes start and theiractivity reaches its peak at optimum temperatures. At too high temperatures, the bio-logical processes will cease (Morison and Butterfield 1990). Yield potential primarilydepends on the amount of solar radiation, especially during the reproductive andmaturity stages (Yoshida and Parao 1976).

The climatic adaptability of the rice crop, which is adapted to a broad latitudi-nal belt ranging from 40°S in central Argentina to 53°N in northeast China, is differ-ent from that of other crops. The rice crop is very sensitive to water deficit during thereproductive to heading stage, resulting in high sterility (Seshu 1989). On the otherhand, extreme temperatures are destructive for rice crop growth and development.The critical low and high temperatures for rice, normally below 20 °C and above 30°C, vary among growth stages (Yoshida 1981). According to Yoshida (1973), highertemperatures result in increased tiller number when light is adequate and low tem-peratures may result in the production of more tillers under low light conditions.

Nishiyama (1976) stated that 9–16 °C and 33 °C are the lower and higher criti-cal temperatures for tiller production and the 25–31 °C range is optimum for tillering.In eastern India, temperatures are mostly within this optimum range and hence tem-perature does not affect tiller production. Chaudhary and Ghildayal (1970), whileworking in eastern India, stated that 10 °C is the lowest critical temperature for tillerproduction. Owen (1972), however, reported that a temperature range of 20–25 °C isoptimum for flower bud initiation in cultivar IR8, whereas 15 °C prevented floralinitiation. Such conditions often arise in eastern Uttar Pradesh, Bihar, and some partsof the northeastern states.

Murata (1976) developed a regression equation between grain yield (Y) anddry weight at heading (Wo) and average solar radiation during ripening and opinedthat Wo is a better measure than solar radiation to assess yield potential. Yoshida andParao (1976) obtained a positive correlation between solar radiation and biomassproduction during the reproductive stage. In eastern India, during good rainfall years,solar radiation becomes a limiting factor for increased biomass and thereby grainyield.

Thus, for the rice crop, all three components of climate—moisture, thermal,and light regimes—individually and combined affect growth, development, and grainyield. Besides climate, other environmental factors such as soil physical and chemi-cal properties, weed flora, and other soil microorganisms vary. All these other envi-ronmental factors in some way or another are influenced by climatic factors. It istherefore necessary to make a thorough inventory of these climatic factors for thecharacterization of different rainfed rice-based cropping systems, which helps in plan-ning agricultural development in any given region.


The eastern India scenario

In eastern India, consisting of Bihar, Orissa, eastern Madhya Pradesh, eastern UttarPradesh, West Bengal, and the northeastern states (Fig. 1), rice is the major cropgrown in favorable as well as adverse and fragile environments. Except in the irri-gated area, which is very small, most of the region is monocropped and rice is grownunder rainfed conditions. The rainfed rice ecosystem in eastern India is again dividedinto different subecosystems such as upland, lowland, and flood-prone. The lowlandscan be further subdivided into favorable and drought-prone.

These subecosystems, in relation to other cropping and production systems,form the overall agricultural scenario of eastern India. In lowlands, a typical cultiva-tion system called the “broadcast beushening” system is adopted, under which riceseeds are broadcast in preplowed fields immediately after the onset of monsoonalrains around 15–30 June. After about 30–35 d, when sufficient water is impounded inthe diked rice fields, the fields are plowed in the standing rice crop. This operation iscalled beushening. Beushening helps (1) reduce the plant population, (2) weed thefields, and (3) create a semipuddled condition to minimize percolation losses to someextent.

In irrigated areas and also in a small portion of rainfed areas, transplanting isbeing practiced. In uplands and in light soils, direct seeding through drilling in lines isalso practiced. The broadcast beushening method, however, is widely adopted.

Sowing, beushening, and later crop operations depend entirely on the vagariesof monsoonal rains. In addition, winter conditions in certain states of eastern India set

Fig. 1. Rainfed rice-growing areas of eastern India.

DibrubarhBahraich

LUCKNOW Gorakhpur

KanpurFatehpur

Allahabad

JabalpurAmbikapur

Raigarh

Raipur

Jagdalpur

Gopalpur

TitlagarhBHUBANESHWAR

CuttackSambalpur

CALCUTTA

PuruliaBurdwan

PATNA

DarbhangaMuzaffarpur

Gaya

Daltenganj Hazaribagh

Ranchi

Jamshedpur

Darjiling

Cherrapunji

GUWAHATI

TezpurJorhat

Silchar IMPHAL

AizawlAgartala


in early, causing damage from low temperature, such as high sterility in the rice crop.As all of eastern India comes under the influence of the southwest monsoon (June-October), solar radiation also becomes a limiting factor during the vegetative stageand early parts of the reproductive stage of the rice crop (mostly in July and August).

Temporal dynamics

Figure 2A-C shows the temporal dynamics of rainfall, maximum and minimum tem-peratures, and solar radiation for three representative stations—coastal, higher lati-tude, and high-rainfall areas—in eastern India. In coastal areas (Fig. 2A), the maxi-mum temperature is always near or above 30 °C and the minimum temperature isalways above 15 °C. In higher latitudes such as in Patna, the minimum temperaturefalls sharply from October onward (Fig. 2B) and this, as stated earlier, may createproblems in grain filling, if these conditions start early. On the other hand, in high-rainfall areas such as Silchar in the northeastern states, with higher rainfall from Mayto September, solar radiation is lower than in other places (Fig. 2C). The minimumtemperature falls below 15 °C only in November. As the rainfall starts early withhigher amounts and intensity, submergence or flood is the main problem in this area,especially in June and July, besides lower amounts of solar radiation.

In view of the importance of these climatic factors, which limit the optimumproductivity of the rice crop, this chapter attempts to analyze three agroclimatic fac-tors—moisture, thermal, and radiation regimes—in eastern India during different stagesof crop growth.


The climatic data required for the analysis are collected from the India Meteorologi-cal Department (IMD). The rainfall probabilities are collected from the data pub-lished by IMD (1995). The solar radiation data required for the analysis are estimated

Fig. 2. Temporal dynamics of rainfall, maximum and minimum temperature, and solarradiation in coastal areas (Cuttack, Orissa, A), higher latitude areas (Patna, Bihar, B),and high-rainfall areas (Silchar, Assam, C).

Month

Max. temp. (°C) Min. temp. (°C) SR Rainfall (mm)

Temperature (°C) and solar radiation (SR)70

50

30

100

J F M A M J J A S O N D

A


B700

500

300

1000


Rainfall (mm)

C


from cloud cover data based on Penman’s (1948) equation. The desirable rainfallperiods are graphically extrapolated as per the criteria given in the text.

The maps are initially prepared by cartographic method and then computerizedthrough scanning. The computerized maps are analyzed through GIS software IDRISI.

Moisture regimeRainfall amount. In eastern India, a major portion of rainfall is received during thesouthwest monsoon season (June-October). In all the states, however, especially inthe northeastern states, rainfall also occurs because of premonsoon thunderstorm ac-tivity during April and May. Figure 3 shows the mean annual rainfall pattern in east-ern India. Annual rainfall varies from more than 3,000 mm in the northeastern statesto less than 1,000 mm in the southern parts of Orissa and eastern Uttar Pradesh. Thehigher amounts of rainfall cause floods in the northeastern states, whereas loweramounts of rainfall cause acute drought conditions in Bihar, Orissa, eastern MadhyaPradesh, and eastern Uttar Pradesh. Using a simple measure of annual rainfall, therainfed rice environments can be characterized. For example, eastern India can bedivided into eight rainfall zones according to rainfed rice cultivation (Table 1).

Fig. 3. Mean annual rainfall pattern in eastern India.

Gonda

Jabalpur

Raipur

Titlagarh

Sambalpur

Cuttack

Balasore

CalcuttaJamshedpurChampaAmbikapur

Gopalpur

Berhampore

HazaribaghRanchi Aizawl

>3,000 mmExtremely high rainfall

2,000–3,000 mmVery high rainfall

1,800–2,000 mmHigh rainfall

1,600–1,800 mmModerately high rainfall

1,400–1,600 mmMedium rainfall

1,200–1,400 mmModerately medium rainfall

1,000–1,200 mmLow rainfall

<1,000 mmExtremely low rainfall


The area in the first three zones is flood- and submergence-prone for longerperiods, whereas the area in the last three zones is drought-prone, with either inter-mittent or terminal drought conditions. The other two zones—moderately high andmedium rainfall—are often favorable for rice cultivation. Even the transplantingmethod of cultivation is followed in this area under rainfed conditions, as mentionedearlier.

Rainy days. Rainy days, in a broader sense, determine crop duration as well ascropping intensity. In eastern India, the number of rainy days varies from more than130 in the northeastern states to less than 50 in eastern Uttar Pradesh (Fig. 4). In amajor portion of eastern India, the annual number of rainy days varies from 50 to 80.

With intermittent dry spells, the crop-growing season is always longer than therainy season. The length of the growing season depends not only on the number ofrainy days or duration of the rainy season but also on moisture availability periodsand soil physical characteristics such as water-holding capacity. Thus, to assess thepossibility of increasing cropping intensity, moisture availability periods are a bettermeasure than just rainy days.

To analyze moisture availability periods, data on the temporal distribution ofrainfall and potential evapotranspiration (PET) are needed. PET values for 44 meteo-rological stations in eastern India are computed using Penman’s (1948) equation.Using PET values, the moisture availability periods for different stations in easternIndia are worked out as described below.

Moisture availability periods. Cocheme and Franquin (1969) designated differ-ent moisture availability periods using rainfall (R) and PET values. Moisture avail-ability periods are categorized as follows:

Moisture availability period Status of rainfall (R) and PET

Humid R > PETMoist PET > R > PET/2Submoist PET/2 > R > PET/4

Table 1. Characterization of rainfall zones in east-ern India.

Annual rainfall (mm) Classification zone

>3000 Extremely high2,000–3,000 Very high1,800–2,000 High1,600–1,800 Moderately high1,400–1,600 Medium1,200–1,400 Moderately medium1,000–1,200 Low<1,000 Extremely low


For rainfed rice cultivation, the submoist period is not suitable and hence theduration of the humid and moist periods alone is discussed in this chapter. The submoistperiod is helpful, however, for the second crop after rice, which is mostly lathyrus,linseed, or chickpea sown as a relay crop in heavy soils with higher moisture reten-tion capacity. The moist period preceding and succeeding the humid period is desig-nated as the moist I and moist II period, respectively. The moist I period is the periodat the beginning of the rainy season when PET values are higher than rainfall. Mostcrop establishment operations are done during this period. The moist II period is theperiod at the end of the rainy season. It is a suitable period for rice crop maturity andharvesting. Also, a relay crop of lathyrus, linseed, or chickpea is sown during thisperiod in the standing rice crop.

Based on these criteria, the moisture availability periods for the 44 meteoro-logical stations in eastern India have been worked out based on the graphical interpo-lation of temporal dynamics of rainfall and PET. Considerable variability exists in themoisture availability periods across the stations in each state. In the northeastern states,the humid period is as high as 278 d in Cherapunji (Meghalaya), whereas in Kanpur(Uttar Pradesh) it is as low as 81 d. Similarly, the moist I period, which is suitable forfield preparation and sowing, starts as early as 2 February in Assam to as late as 9 July

Fig. 4. Annual number of rainy days in eastern India.

Gopalpur

Jagdalpur

Kanker Titlagarh

Berhampore

Raipur Sambalpur

Cuttack

JabalpurPendro

Raigarh

Champa

AmbikapurUherte

BalasoreCalcutta

Midnapur

Jamshedpur

BurdwanPurulia

Ranchi

HazaribaghGaya

DaltenganjSafna

Allahabad

Gonda

GorakhpurDarbhanga

PatnaBaghalpur

Sabour

BoghleraDhuburi

Aizawl

100–120

120–130

>130

<50

50–60

60–70

70–80

<80

<90

90–100

Fatehpur

KanpurLucknow

Bahraich


in eastern Uttar Pradesh under normal conditions. Table 2 shows the variations insowing periods (moist I period) in different states of eastern India. In general, theentire sowing operation in eastern India should be completed within 20–25 d.

Within a classified rainfall zone, significant variations occur in the total dura-tion of moisture availability periods across locations (Table 3). Within the rainfallzone of 1,200–1,400 mm, the humid period is 95 d in Gaya and 128 d in Sabour(Bihar). Similarly, in the rainfall zone of 1,400–1,600 mm, the humid period is 153 din Cuttack (Orissa) and only 117 d in Raipur (Madhya Pradesh). This clearly impliesthat varietal duration should be based on the duration of the moisture availabilityperiod but not on the amount of rainfall.

Table 2. Range of the beginning of the sowing period (beginning ofmoist period I) in different states in eastern India.

Sowing periodState

Variation Duration (d)

Eastern Madhya Pradesh 31 May–21 Jun 21Eastern Uttar Pradesh 11 Jun–9 Jul 28Bihar 5 Jun–7 Jul 32Orissa 19 May–9 Jun 20West Bengal 28 Apr–9 Jun 42Assam 2 Feb–28 Mar 54Meghalaya 14 Feb–6 Mar 20Mizoram 22 Mar–18 Apr 27

Table 3. Variation in moisture availability periods of two different rainfall zones.

Moisture availabilitya

AnnualSite rainfall Moist I Humid Moist II

(mm)D P D P D P

Gaya, Bihar 1,200–1,400 17 15 Jun–10 Jul 95 11 Jul–13 Oct 22 14 Oct–10 Nov

Sabour, Bihar 1,200–1,400 16 7–22 Jun 128 23 Jun–28 Oct 18 29 Oct–15 Nov

Raipur, 1,400–1,600 14 16–29 Jun 117 30 Jun–24 Oct 18 25 Oct–11 NovMadhyaPradesh

Cuttack, 1,400–1,600 18 30 May–17 Jun 153 18 Jun–17 Nov 18 18 Nov–5 DecOrissa

aD = duration in days, P = period.


Table 4 shows the range of different moisture availability periods in each stateof eastern India. The duration of the moist I period varies from 10 to 33 d in differentstates, whereas that of the moist II period varies from 9 to 22 d. The humid period,which is very important for the rice crop’s growth and development, varies from 81 to123 d in Uttar Pradesh to 224 to 278 d in the northeastern states. The higher numberof days under the humid period also includes the submergence (flooding) period insome states. Those areas with a longer humid period duration are suitable for increas-ing cropping intensity. The kind of crop to be grown again depends on the soil typeand thermal regime during the winter period (November-March).

Rainfall probabilities. The discussion so far is based on average rainfall condi-tions. But to improve the productivity of rice in eastern India, especially under rainfedconditions, it is more realistic if the rainfall analysis is carried out at different prob-ability levels and interpreted accordingly. The India Meteorological Department (IMD1995) published the probabilities of different quantities of weekly rainfall for all themeteorological stations in India. The data for the period 20th to 45th standard meteo-rological week (May to October) have been used to analyze and characterize rainfedrice environments in different states in eastern India. The average weekly rainfall formost of the stations in eastern India matched with either 30% or 40% probability.This implies that any agricultural planning on the basis of average rainfall data has aprobability of success once in three years only and, in the other two years, rainfall isalways below average. This may affect rice crop growth and development, especiallyin upland and lowland drought-prone ecosystems. This clearly indicates that, in east-ern India, rainfed rice-growing environments can be characterized in a better way byconsidering rainfall probabilities, especially desirable rainfall probabilities.

Desirable rainfall for rainfed rice cultivation. The average daily water lossesby potential evapotranspiration and percolation in rice fields of eastern India accountfor approximately 3 to 4 mm each. Therefore, the total water requirement for waterstress-free rice cultivation under rainfed conditions can be assured at a minimum of 7mm d–1. Also, by considering a minimum value of 7 mm d–1, it is imperative that the

Table 4. Range of moisture availability periods in different states ofeastern India.

Range of moisture availability periods (d)State

Moist I Humid Moist II

Madhya Pradesh 10–16 103–147 12–21Uttar Pradesh 12–18 81–123 17–22Bihar 12–18 95–139 17–22Orissa 12–24 131–165 15–18West Bengal 16–28 130–208 11–22Assam 25–33 191–270 17–22Other northeastern states 21–28 224–278 9–17


fluctuations/variations in rainfall above this threshold value do not influence cropgrowth and development.

Thus, considering desirable rainfall for rainfed rice cultivation to be a mini-mum of 7 mm d–1, the duration and starting periods of desirable rainfall for averageand different probabilities are worked out. The important features of the analysisfollow.

Figure 5 shows the duration of desirable rainfall under average conditions ineastern India, which varies from less than 60 d in a small pocket in eastern UttarPradesh to more than 200 d in Meghalaya. In a major portion of eastern India (exclud-ing the northeastern states), the desirable rainfall period is around 80 to 100 d, whilein the northeastern states the average duration varies from 140 to 200 d. Thus, theproblem in eastern India (excluding the northeastern states) is water stress at onestage or another, whereas in the northeastern states excess moisture or submergencelimits rice productivity. In Berhampore District of West Bengal, the desirable rainfallperiod occurs in two splits, implying that in some parts of West Bengal submergenceand drought can occur in the same cropping season. Such factors need to be thor-oughly examined for developing improvement strategies for rice productivity at thedistrict or microregional level, such as tehsil or blocks.

Fig. 5. Duration (in days) of desirable rainfall periods in eastern India.

>100

90–100

80–90

70–80

60–70

>60

>200

180–200

Kanker

Raipur

Jabalpur

Uherte

Calcutta

Gaya

Battonganj

Gorkhore

Safna

Lucknow

Barchie

Dibrubarh

JorhatTezpur

GUWAHATI

Cherrapunji

DhuburiDarjiling

PATNA

DarbhangaMuzaffarpur

GorakhpurGonda

LUCKNOW

Bahraich

KanpurFatehpur

Allahabad

SetnaUmaria

Jabalpur

Ralpur

Kanker

Jagdalpur

Gopalpur

Titlagarh

BHUBANESHWAR

CuttackSambalpur

JharsugudaRaigarh

Champa

Ambikapur

Pendro

Belasore

CALCUTTA

Midnapur

Purulia Burdwan

Gaya

Daltenganj Sabour

Hazaribagh

Ranchi

Baghalpur

BerthamporeSilchar

IMPHAL

160–180

140–160

120–140

>120


A knowledge of the starting of desirable rainfall periods, besides their duration,helps in developing crop calendars to suit a particular location, which help in plan-ning different crop operations in the growing season. Hence, the starting dates foraverage and different probabilities of desirable rainfall periods have been worked outand are discussed below.

The analysis of average dates of starting periods of desirable rainfall in easternIndia revealed that, in the northeastern states, desirable rainfall starts as early as theend of March in Manipur and from 1 to 5 April in a major portion of the northeasternstates. In Assam, the desirable rainfall period starts on or just after 15 April.

In other eastern states, the starting period varies from 15 June to 5 July. It has adirect relation with the onset of the southwest monsoon, but with a slight lag of about10 to 15 d. The average dates of the start of the desirable rainfall period are around 15June in Orissa and around 5 July in the western parts of eastern Uttar Pradesh. Theearly start of the desirable rainfall period in the northeastern states clearly shows thedependability of rice cultivation on the premonsoon thunderstorm rainfall for cropoperations.

Analysis of desirable rainfall periods at different probabilities: a summary bystates. The results of the analysis of desirable rainfall periods for all the stations ineach state are discussed in detail below.

In Orissa, the average duration of desirable rainfall varies from 33 d at Gopalpurto 120 d at Balasore. At Gopalpur, the average duration is not only less but also notcontinuous. This implies that in this area rice cultivation without supplemental irriga-tion is not possible. At 50% and higher probability, duration is either too low or nil atall the places in Orissa except Sambalpur, where duration of the desirable rainfallperiod is 81, 65, and 51 d at 50%, 60%, and 70% probabilities, respectively. Thisindicates that there are few opportunities for rainfed rice cultivation without waterstress in most of these places.

For eastern Madhya Pradesh, the desirable rainfall period at 60% probability ismore than 60 d except at Raipur and Satna. This is a clear indication of favorableconditions for rainfed rice cultivation. In these districts with proper water manage-ment practices, rice productivity can be improved considerably because other condi-tions are favorable.

In Bihar, with considerable area under uplands, the average duration of desir-able rainfall periods varies from 67 d in Baghalpur to 97 d in Ranchi. At Baghalpur,the desirable rainfall period starts later (1 July) than in other areas. At 50% probabil-ity, there is no desirable rainfall period at this station, whereas at other stations dura-tion varies from 36 to 73 d with a discontinuity in between at Darbhanga and Gaya.Ranchi has 54 d of desirable rainfall at 60% probability, indicating a better scope forimproving rice productivity by adopting better water management practices.

In eastern Uttar Pradesh, the average duration of the desirable rainfall periodvaries from 63 d at Allahabad to 93 d at Bahraich. At Bahraich, however, the durationof desirable rainfall decreases sharply at 40% and 50% probability. This shows thatBahraich has a higher variability of rainfall and less dependability; hence, rice pro-ductivity is very low at Bahraich compared with other parts of Uttar Pradesh. Also,


the desirable rainfall period at 50% probability at all stations in eastern Uttar Pradeshends by mid-August. Thus, terminal drought is a recurring feature in eastern UttarPradesh.

In West Bengal, the average duration of the desirable rainfall period is around100 d but it starts earlier than in other eastern states. Duration is less, however, whencompared with eastern Madhya Pradesh and Orissa. At 50% probability, durationvaries from 39 to 57 d in different parts of West Bengal. In this state, however, theintensity of rainfall is higher, as the rains are received because of cyclonic activity inthe Bay of Bengal, resulting in flooding in rice fields; hence, deepwater rice is moreprevalent.

For the northeastern states, the desirable rainfall situation is quite different fromthat of the other states of eastern India. The average duration is high and starts about50 to 60 d earlier than in other states. At 60% probability, duration varies from as highas 145 d at Silchar to as low as 3 d at Guwahati. Besides duration, the intensity ofrainfall is also high and the terrain is undulated in this area. Therefore, floods andsometimes flash floods are a recurring feature.

It is therefore necessary to analyze rainfall and other agroecological character-istics of the northeastern states with a different approach. The strategies for improv-ing rice productivity in these states must be different from those of other states ofeastern India.

Thermal regimeIn eastern India, in general, the thermal regime (both the maximum and minimumtemperatures) is favorable for rice cultivation. Higher day temperatures at the initialstages and lower night temperatures during the anthesis period, however, sometimeslimit the crop’s growth and development. In some high-rainfall areas, rice is grownunder rainfed conditions from September-October either after the floodwater recedesor after the harvest of the first crop sown in April-May. For this crop, the temperaturesin the winter (November-March) season are also important. In view of this, the maxi-mum and minimum temperatures of both the rainy (June-October) and winter (No-vember-Dececember) seasons are analyzed and discussed below.

Maximum temperatures. The maximum temperature during the rainy seasonranges mostly between 30 and 34 °C. Higher day temperatures prevail in the western-most part of eastern Uttar Pradesh, where the onset of monsoon is relatively late andtherefore, in some years with its late onset (after 1 July), higher day temperaturesduring the sowing/germination time may affect the crop. In the rest of the region,maximum temperatures are favorable during the rainy season.

Minimum temperatures. The minimum temperature during the rainy seasonvaries little throughout eastern India. In eastern Uttar Pradesh, Bihar, West Bengal,and coastal Orissa, it is more than 25 °C, whereas, in a majority of the area in thenortheastern states and in eastern Madhya Pradesh, it varies from 23 to 25 °C. In asmall area in eastern Madhya Pradesh and Meghalaya, it is less than 23 °C.

The thermal regime in general is favorable for rice cultivation in eastern India.As mentioned earlier, however, temperature becomes a limiting factor only when


winter conditions set in early and the night temperature falls below 15 °C at the grain-filling stage, as occurs in Bihar and eastern Uttar Pradesh. Figure 6 shows the averageweekly temperature pattern at Raipur and cardinal temperatures at various phenologi-cal stages. The maximum temperature during the vegetative stage is slightly belowthe optimum temperature limits (Mavi 1976) . Similarly, during the grain-filling stage,the minimum temperature, which limits grain filling, is above the threshold values(Fig. 6) under average conditions at Raipur. Even under such favorable conditions,the minimum temperature sometimes falls below the threshold limits, thus causingsterility problems. Variations also occur in varietal interaction with such low nighttemperatures. A detailed analysis is needed to examine varietal interaction with lownight temperatures.

Radiation regimeAs mentioned earlier, radiation is a limiting factor for rice in the vegetative and earlyreproductive stages of crop growth as eastern India comes under the influence of thesouthwest monsoon. Even during a dry period, radiation values are lower in this re-gion as the sky is mostly overcast. Radiation is therefore a very important climaticcomponent for rice cultivation in this region during the rainy season.

In India, the network of radiation measuring stations is sparse. Data on cloudamount in octos measured twice a day, however, are readily available. Therefore,radiation values were estimated for all 44 meteorological stations in eastern Indiausing Penman’s (1948) equation.

Fig. 6. Pattern of weekly averages of maximum and minimum tem-peratures in a normal year and in 1985 during the rice-growing sea-son compared with the optimum values in different growth stages.

Maximum (normal)Maximum (1985)Minimum (normal)Minimum (1985)

50

40

30

20

10

0

Temperature (°C)

Ger

min

atio

n

See

dlin

g

Vege

tativ

e

Pani

cle

initi

atio

n

Gra

in f

illin

g

Mat

urity

0 2 4 6 8 10 12 14 16 18 20

Weeks from sowing

20°20°

22°

31°30°30°


The radiation values range from less than 12.5 to more than 16.5 MJ m–2 d–1 inJanuary and they start increasing thereafter until May. In May, the radiation valuesreach their peak at more than 25.0 MJ m–2 d–1 in eastern Uttar Pradesh. In the north-eastern states, however, the peak values are reached in April and decrease thereafter,as premonsoon and then monsoonal rains start. The solar radiation values in easternIndia are lowest in July and August and range from less than 8.5 to slightly more than12.5 MJ m–2 d–1 (Fig. 7A,B). For better crop growth and development, a minimumvalue of 12.0 MJ m–2 d–1 is needed.

Figure 2 also shows the temporal dynamics of solar radiation at three represen-tative stations. At Cuttack, where the monsoon starts earlier (mid-June), the solarradiation is less than 12.0 MJ m–2 d–1 from July to September, whereas, in Patna,where the monsoon starts late, in the first week of July, solar radiation values are lessthan 12.0 MJ m–2 d–1 only in August and September. In the rest of the months, thevalues are more than 12.0 MJ m–2 d–1. In the case of Silchar, however, the solarradiation values are lower than 12 MJ m–2 d–1 from June to September. This is a clearindication that, in higher rainfall areas, solar radiation constrains the rice crop’s growthand development.

Summary and conclusions

To develop strategies for improving rice productivity, a thorough inventory of thethree components of climate—moisture, thermal, and radiation regimes—is needed.In eastern India, rice is grown mostly under rainfed conditions in three subecosystems:rainfed lowland, upland, and flood-prone. In this region, rainfall or moisture regimeis an important climatic component because most of the area is under upland andrainfed lowland drought-prone ecosystems. Rainfall amount and rainy days are simplemeasures for characterizing rainfed rice environments. Using rainfall amounts, east-ern India is divided into eight zones, of which three are under flood/submergence-prone conditions and three are under drought-prone conditions. Only two zones comeunder favorable conditions. In both the flood-prone and drought-prone areas, betterwater management practices need to be developed.

Rainy days, though readily available, are not a good measure for assessing thegrowing season as the growing season is always longer than rainy days due to inter-mittent dry spells. Therefore, moisture availability periods are a relatively better mea-sure for assessing the crop-growing season and cropping pattern. The humid period(P > PE) is the best period for the rice crop’s vegetative and reproductive phases;therefore, varietal duration needs to be assessed based on the humid period. The rice-based second crop in the winter season can be grown in areas with a longer durationof humid and moist II periods and in soils with a higher moisture retention capacity.

In spite of such detailed analysis, in eastern India average rainfall matches withonly 30% or 40% probability, implying that any strategy based on average rainfallcan be successful only once in three years. Hence, a concept of desirable rainfallperiod has been developed by defining a desirable rainfall period as the period whendaily rainfall is 7 mm or more. The rainfall fluctuations above this threshold value do


Fig. 7. Solar radiation (MJ m–2 d–1) in July (A) and August (B) over eastern India.

30°N

25

20

1580 85 90 95 100°E75

Gopalpur

Raipur Sambalpur

Titlagarh

Cuttack

Jabalpur

B

GondaLucknow

JabalpurPendro

Raipur

Gopalpur

JagdalpurBerhampore

Champa

Cuttack

SambalpurBalasore

MidnapurCalcutta

>12.5 8.5–10.5

10.5–12.5 <8.5

30°N

25

20

15

A


not affect the growth and development of rainfed rice. The analysis of the desirablerainfall periods at different probabilities, however, indicates that the duration of thedesirable rainfall period at 60% or higher probability is very low in most states ineastern India. This suggests the need to develop better water management practices toimprove rice productivity.

An analysis of the thermal regime indicated that, in general, it is favorable forrainfed rice cultivation in eastern India under normal conditions. In some years, how-ever, when winter conditions set in early, low night temperature may lead to sterilityconditions in rice.

In eastern India, radiation is a limiting factor during the vegetative and repro-ductive stages. With a threshold value of 12.0 MJ m–2 d–1 as the lower limit (approxi-mately 300 calories m–2 d–1), radiation is more constant in high-rainfall areas than inlow-rainfall areas of eastern India.

Thus, a thorough inventory of the three climatic components revealed that, inmost of the places, rainwater management in both flood-prone and drought-proneareas is the most important strategy for increasing rainfed rice productivity. Watermanagement practices include better drainage in flood-prone areas and rainwater har-vesting and recycling in drought-prone areas. The thermal regime is more or lessfavorable, but, for radiation, it is necessary to identify suitable varieties with higherenergy-use efficiency.

ReferencesChaudhary TN, Ghildayal BP. 1970. Influence of submerged soil temperature regimes on growth,

yield and nutrient composition of the rice plant. Agron. J. 62:282-285.Cocheme J, Franquin P. 1969. An agroclimate survey of a semi-arid area in Africa-South of

Sahara. FAO/WMO Technical Bulletin 86.Frankel OH. 1976. The IRRI phytotron: science in the service of human welfare. In: Climate

and rice. Los Baños (Philippines): International Rice Research Institute. p 3-9.IMD (India Meteorological Department). 1995. Weekly rainfall probability for selected sta-

tions of India. Vol II. Pune (India): IMD. 517 p.Mavi HS. 1976. Introduction to agrometeorology. New Delhi (India): Oxford & IBH Publish-

ing Co. 237 p.Morison JLL, Butterfield RE. 1990. Cereal crop damage by frosts; spring 1990. Weather

45(8):308-317.Murata Y. 1976. Productivity of rice in different climatic regions of Japan. In: Climate and rice.

Los Baños (Philippines): International Rice Research Institute. p 449-470.Nishiyama I. 1976. Effect of temperature on the vegetative growth of rice. In: Climate and rice.

Los Baños (Philippines): International Rice Research Institute. p 159-185.Owen PC. 1972. Effects of night temperature on growth and development of IR8 rice. Exp.

Agric. 8:213-218.Penman HL. 1948. Natural evaporation from open water bare soil and grasses. Proc. R. Soc.

(A) 193. 120 p.Sastry PSN. 1976. Climate and crop planning with particular reference to rainfall. In: Climate

and rice. Los Baños (Philippines): International Rice Research Institute. p 51-63.


Seshu DV. 1989. Impact of major weather factors on rice production. In: Agrometeorologicalinformation for planning and operation in agriculture with particular reference to plantprotection. Geneva (Switzerland). p 41-63.

Yoshida S. 1973. Effect of temperature on growth of the rice plant (Oryza sativa L.) in con-trolled environment. Soil Sci. Plant Nutr. 19:299-310.

Yoshida S. 1981. Fundamentals of rice crop science. Los Baños (Philippines): InternationalRice Research Institute. 269 p.

Yoshida S, Parao FT. 1976. Climatic influence on the yield and yield components of lowlandrice in the tropics. In: Climate and rice. Los Baños (Philippines): International RiceResearch Institute. p 471-494.

NotesAuthors’ addresses: A.S.R.A.S. Sastri, Indira Gandhi Agricultural University, Raipur, India;

V.P. Singh, International Rice Research Institute, Los Baños, Philippines.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-


Agrohydrologic and drought risk analyses of rainfed cultivation . . . 233

Drought is a common problem in the northwest region of Bangladesh, wherethe monsoon season (June-October) receives only about 1,000 mm of rain-fall. For drought characterization, long-term (1961-93) weather data wereanalyzed and the impact of drought on rice establishment and farmers’ man-agement practices was studied for two wet seasons (1994-95). The prob-abilities of two and three consecutive 5-d droughts occurring during the grain-filling period of transplanted rice (TPR) are 73% and 53%, respectively. In anaverage year, rainfall may be adequate for transplanting by mid-July. But,twice in ten years, the required rainfall may not be available by 15 Augustand transplanting may be delayed. The average seasonal relative water sup-ply (RWS) from rainfall is 0.79. Because of late transplanting, the RWS isexpected to be only about 0.51, twice in ten years, and can be very detrimen-tal to crop yield. Dry-seeded rice (DSR) may be established early, by the firstweek of June in an average year, and, twice in ten years, by the third week ofJune. DSR yields are similar to those of TPR, but DSR matured 1–2 wk earlierthan TPR and left a better soil-water regime for the subsequent nonrice crop.

In Bangladesh, more than 50% of the rice areas are rainfed lowland (IRRI 1993) andmost of these areas suffer from either flood or drought. In flood-prone areas, farmersminimize yield losses by selecting crops in accordance with flood depth, duration,and timing. In drought-prone areas, however, farmers often ignore the possibilities ofdrought and sometimes suffer significant yield losses. MPO (1985) estimated that theaverage yield reduction between irrigated and rainfed (drought-prone) situations formodern variety (MV) aus (premonsoon) rice was 23% and for MV aman (monsoon)rice was 31%.

Drought effects are most severe in the northwest region of the country, espe-cially in the Barind Tract, which covers about 0.16 million ha, of which nearly 0.1million ha is rainfed. The average seasonal rainfall of about 1,000 mm during the fivemonsoon months (June-October) is the lowest in the country (Manalo 1977) and can

Agrohydrologic and drought riskanalyses of rainfed cultivationin northwest BangladeshA.F.M. Saleh, M.A. Mazid, and S.I. Bhuiyan

234 Saleh et al

be classified as low for a rainfed ecosystem (Garrity et al 1986). The situation isexacerbated by drought spells of 1 or 2 wk within a month, which, depending on theirtiming, can drastically affect crop yield.

As both technical and economic constraints limit further irrigation develop-ment in Bangladesh, the majority of the farmers in rainfed systems will certainlycontinue to remain in the drought-prone environment in the foreseeable future. Tech-nologies for drought alleviation are now available and are gradually being acceptedand adopted by farmers in rainfed lowlands. Drought alleviation is possible by switch-ing from the traditional transplanting method of crop establishment (TPR) to dryseeding (DSR) in which rice seeds are sown on dry-tilled unsaturated fields early inthe season. These fields may subsequently become flooded with the onset of the mon-soon rains. Studies on DSR in the Philippines (Saleh and Bhuiyan 1995, Lantican etal 1999) have shown that DSR uses rainfall more efficiently and suffers less droughtrisk than TPR. On-farm reservoirs for supplementary irrigation have also been usedeffectively for drought alleviation in the Philippines (Moya et al 1986) and in India(Paul and Tiwari 1994). But, before planning and recommending such interventionsfor drought alleviation in a rainfed ecosystem, a systematic and quantitative analysisof drought is imperative.

This study is an attempt in that direction with the specific objectives of (1)characterizing the area in terms of agrohydrology, (2) studying the nature, extent, andfrequency of droughts during the aman season from long-term rainfall data, and (3)studying the relative merits of dry seeding over transplanting of rice in terms of droughtalleviation.


Agrohydrologic settingThe field study was carried out at Rajabari Union of Godagari Thana of RajshahiDistrict, in the Barind Tract of northwest Bangladesh. The Barind Tract has an undu-lating topography with gray terrace soil and an average elevation of 43 m above sealevel. The soil texture varies from silt loam to silty clay loam and is poorly drainedwith a 6–8-cm thick plow pan at 9–11-cm depth (Mazid et al 1998). The soil is low inorganic matter (0.8–1.2%) and is acidic (pH from 5.5 to 6.5).

Annual rainfall at Godagari averages about 1,300 mm and about 80% of it isconfined to the monsoon months of June to September. The average daily evapora-tion at Rajshahi, located centrally within the Barind Tract, varies from 2.3 mm inJanuary to 6.3 mm in April. The four months when the rainfall exceeds evaporationare June to September. The maximum and minimum temperatures at Rajshahi aver-age about 39 °C and 10 °C, respectively, and occur in April and January.

The predominant cropping pattern in the area is transplanted aman rice–fallow.In some areas, chickpea, linseed, barley, and other crops are grown either as a monocropor as a mixed crop during the rabi (dry) season.


Analytical procedureThe nature and extent of droughts in the area were studied by using the water balancemethod to find out the numbers of 5-d water-deficient periods and their continuity(consecutive 5-d periods) during the aman season. In these periods, the rainfall wasinadequate to meet the crop water requirement. Since rainfall during one day may beadequate for the entire period considered, the chances of a drought within the periodare higher if the period is longer than 5 d. As soil water may be available to the riceroots 2 to 3 d after standing water has disappeared, a shorter duration was not consid-ered. The methodology followed is similar to Thornthwaite’s method of water bal-ance (Steenhuis and Van der Molle 1986, Paul and Tiwari 1992). In this analysis,however, the seepage and percolation loss (S&P) was also taken into account alongwith evapotranspiration (ET), as S&P is an integral part of the water requirement forlowland rice. The 5-d water balance is written as follows:

Ht = Rt + Ht–1 – ETt – (S&P)t (1)

where H is bund storage, R is rainfall, and subscripts t and t–1 denote time in 5-d timesteps (present and previous 5 d, respectively). R, ET, and S&P are all expressed inmm d–1 (in each 5-d time step). Thus, the incidence of either drought or adequacy ofwater supply was determined by the following criteria:

If (Rt + Ht–1 – ETt ) < 0, then (S&P)t = 0 and Ht = 0; there is drought (2)If ETt < (Rt + Ht–1) < ETt – (S&P)t and then Ht = 0; there is no drought (3)

If Rt + Ht–1 – ETt – (S&P)t > 0 and Ht > 0; there is no drought (4)If Rt + Ht–1 – ETt – (S&P)t > Hmax and Ht = Hmax = 20 cm; there is no drought (5)

Water availability during the crop growth season was determined by using the con-cept of relative water supply (RWS), which is defined as the ratio of water supply todemand by the crop. Thus, RWS for a given period t can be written as

RWSt = [(R/(ET + S&P)]t (6)

An RWS value greater than 1 indicates that the water supply is abundant, whereas avalue less than 1 indicates drought. For lowland rice, pan evaporation data are goodindicators of crop evapotranspiration (Tomar and O’Toole 1980).

Rainfall probability and drought occurrences were analyzed by using gammadistribution (Thom 1968).

For drought analysis, 31 years of daily rainfall data (1963-93) of Godagari sta-tion, situated about 15 km northwest of the study area, were collected from theBangladesh Water Development Board as long-term rainfall data were not availablefor the study site. Daily rainfall and U.S. Class A pan evaporation data for the studyseasons were collected from a nearby temporary weather station. Past crop yield datawere collected from the Bangladesh Bureau of Statistics (1979-83) and the Depart-

236 Saleh et al

ment of Agricultural Extension (1985-93) for studying the impact of droughts on cropyield.

Field experiment and farmers’ surveyThe field study was carried out during the 1994 and 1995 aman (July-November)season. Crop production practices of 50 randomly selected farmers were surveyedduring the study period using a predesigned questionnaire to (1) gain qualitative in-formation on farmers’ practices in crop establishment, crop management, and pro-duction limitations and (2) collect quantitative information on timings of farmingactivities, input use, and productivity of the aman crop.

Field water status and water use of 30 farms were closely monitored by install-ing 100-cm-long PVC tubes (50 cm perforated) 20 cm above the ground surface andsome below the ground surface. The water-level readings in the PVC tubes (standingor perched water level) were taken every other day. Changes in standing-water levelsduring rainless days were used to estimate S&P loss after deducting pan evaporation.

Field experiments on effect of crop establishment method (DSR and TPR) andseeding/transplanting date on crop yield were carried out at Rajabari in a split-plotdesign. For transplanting, seedlings were raised in the wet bed on the same day ofseeding of DSR and then 30-d-old seedlings were transplanted in puddled soil. Thetransplanting schedule was delayed by 15 d in 1994 because of inadequate rainfall forpuddling.


Farming activities in relation to rainfallFigure 1 shows the timing of farming activities for 50 transplanted rice-growing (thetraditional practice) farmers sampled during the 1994 and 1995 aman seasons. Farm-ers began the first plowing and sowing activities earlier in 1994 because of the 60-mm rainfall in April. But transplanting could not be carried out until July, when therewas enough rainfall for land preparation. In 1995, the first plowing and seeding started

Fig. 1. Progress of farming activities during the 1994 (A) and 1995 (B) aman season at Rajabari,Godagari, Rajshahi.

Apr May Jun Jul Aug Sep Oct Nov Dec

500

400

300

200

100

0

Rainfall (mm)

B

Month

Farmers (%)

A

Apr May Jun Jul Aug Sep Oct Nov Dec

100

80

60

40

20

0

First plowSeedingTransplantHarvest


late because of low rainfall at the beginning of the season. Heavy rainfall from thebeginning of June resulted in quick completion of farming activities and transplant-ing was completed by the end of July.

The average time difference between the first plowing and transplanting was 62d in 1994. The duration was only 31 d in 1995 due to more favorable rainfall condi-tions. The average time difference between seeding and transplanting was 45 d in1994 and 40 d in 1995. In both years, 50% of the farmers completed transplanting bymid-July but harvesting was completed by mid-November in 1994 and the end ofNovember in 1995. Heavy rain in mid-November (50 mm as shown in Figure 1)delayed harvesting in 1995. The survey data showed that, for 65% of the farmers, thepreferred times of transplanting and harvesting are mid-July and mid-November. Thus,the 120 d between 15 July and 15 November were considered as the preferred fieldduration of the crop in further analysis. The average crop field duration was 119 d in1994 and 130 d in 1995.

During the two months required from first plowing to transplanting by 50% ofthe farmers (15 May to 15 July), the amount of rainfall in 1994 was 380 mm. Duringthe same period in 1995 (in this case 19 June to 20 July), rainfall was 391 mm. Hence,it was assumed in further analysis that, in the study area, the farmers required at least400 mm of rainfall to complete land preparation before transplanting.

About 87% of the farmers grew modern rice varieties, whose average yield was3.7 t ha–1 and range was 1.8 to 4.7 t ha–1. The average use of fertilizer was high: 100kg N ha–1, 17 kg P ha–1, and 16 kg K ha–1. Nitrogen application in two splits was mostcommon (44%), followed by three splits (36%). About half the farmers applied or-ganic fertilizer (manure) during land preparation at an average rate of 3.5 t ha–1. Aboutone-third of the farmers grew a second crop following the harvest of aman rice. Themajor second crop was chickpea, which was grown by about two-thirds of the farm-ers.

Drought analysisDrought in the study area was characterized through probability analysis of rainfallduring the crop field duration (15 July to 15 November). The 50% (rainfall equal to orexceeding the specified amount in 5 out of 10 years) and 80% dependable (rainfallequal to or exceeding the specified amount in 8 out of 10 years) crop field durationrainfalls are 745 mm and 597 mm, respectively.

Water balance with 5-d time steps was carried out for the crop field durationusing equations 1–4 and with 31 years of rainfall data. The field duration was dividedinto three stages: vegetative, reproductive, and grain filling, each stage being 40 dlong. The average pan evaporation during the season was 3 mm d–1 and the averageS&P loss was 7 mm d–1. The numbers of 5-d drought periods and consecutive 5-ddrought periods during each growth stage for the past 31 years are given in Tables 1and 2. They are most frequent during the grain-filling stage. On average, about seven5-d droughts are expected to occur during crop field duration. But, as these are notcontinuous, they are not expected to seriously affect yield. About three consecutive

238 Saleh et al

Table 2. Number of consecutive 5-d droughts during crop field duration.

CycleYear Seasonal

Vege- Repro- Grain maximumtative ductive filling

1979 1 2 4 71963 0 0 0 01964 0 0 0 01965 0 3 5 51966 0 3 3 61967 0 2 5 71968 0 3 5 51969 0 0 3 31970 3 2 0 31971 0 2 3 31972 0 4 4 41973 2 0 2 21974 0 2 3 31975 2 2 2 21976 0 0 5 51977 0 0 0 01978 2 0 4 41979 0 0 4 4

CycleYear Seasonal

Vege- Repro- Grain maximumtative ductive filling

1979 1 2 4 71980 0 0 0 01981 0 0 5 61982 3 2 5 71983 0 0 3 31984 0 0 2 21985 4 3 0 41986 0 0 5 51987 0 0 5 61988 0 3 4 41989 2 0 2 21990 0 2 3 31991 0 0 5 51992 2 0 2 21993 0 0 0 0

Average 0.7 1.1 2.9 3.4

Table 1. Number of 5-d droughts during crop field duration.

CycleYear Total

Vege- Repro- Graintative ductive filling

1963 1 1 1 31964 0 2 2 41965 3 3 5 111966 1 4 4 91967 2 3 5 101968 2 3 5 101969 1 1 3 51970 3 3 2 81971 1 1 3 51972 2 4 4 101973 4 0 3 71974 2 2 3 71975 2 5 4 111976 1 0 5 61977 2 2 1 51978 2 2 4 81979 1 2 4 7

CycleYear Total

Vege- Repro- Graintative ductive filling

1980 0 3 1 41981 0 3 5 81982 4 3 5 121983 1 1 3 51984 1 1 2 41985 4 4 2 101986 2 1 5 81987 1 1 5 71988 2 3 4 91989 2 1 2 51990 1 2 3 61991 1 0 5 61992 2 2 3 71993 2 0 2 4

Average 1.7 2.0 3.4 7.1


5-d droughts can be expected during the grain-filling stage compared with none dur-ing both the vegetative and reproductive stages.

Figures 2 and 3 show the probability analysis of 5-d droughts and consecutive5-d droughts. The probability of two 5-d droughts (not consecutive) during the grain-filling stage is more than 80%. The probability of 10-d and 15-d droughts (two andthree consecutive 5-d droughts) during the grain-filling stage is 73% and 53%, re-spectively. Thus, a 2-wk period without rain is expected once in two years and can bedetrimental to crop yield.

Figures 4 and 5 show a scatter diagram of seasonal rainfall (15 July-15 Novem-ber) and total numbers of 5-d droughts and consecutive 5-d droughts. There is a faircorrelation (r = 0.56 and significant at the 1% level) between total seasonal rain andtotal number of 5-d droughts. Seasonal rainfall and total number of consecutive 5-ddroughts, however, are uncorrelated.

The 50% and 80% dependable 5-d rainfalls for May-November were deter-mined and are shown in Figure 6, which shows that in the study area little or norainfall is expected after early October. Since early October is the beginning of thegrain-filling stage, the crop is vulnerable to damage by drought.

Figure 7 shows the probability of getting the 400 mm of rainfall required fortransplanting at different times at the beginning of the crop season. In an average year(50% probability), the 400 mm of rainfall required for transplanting can be expectedby 15 July. Twice in ten years (80% probability), however, this amount may not beavailable before 15 August and transplanting may be delayed by one month.

A water adequacy analysis was carried out for crop field duration by calculat-ing the relative water supply using equation 5 in each 5-d period. The average sea-sonal RWS value at 50% probability, when transplanting is completed by 15 July(average year), is 0.79. This means that, even in an average year, the water supplyfrom rainfall is inadequate to meet the crop water requirement. If transplanting isdelayed because of inadequate rainfall at the beginning of the season (completed after

Fig. 2. Probability of 5-d droughts during thecrop season.

Fig. 3. Probability of consecutive 5-d droughtsduring the crop season.

100

80

60

40

20

02 3 4 5 2 3 4 5 2 3 4 5

Probability (%)

Consecutive 5-d drought periods

VegetativeReproductiveGrain filling

100

80

60

40

20

01 2 3 4 5 1 2 3 4 5 1 2 3 4 5

Probability (%)

Number of 5-d drought periods

VegetativeReproductiveGrain filling

240 Saleh et al

Fig. 4. Scatter diagram of seasonal rainfall andtotal 5-d droughts.

Fig. 5. Scatter diagram of seasonal rainfall andtotal consecutive 5-d droughts.

Fig. 7. Cumulative rainfall at 50% and 80% prob-abilities.

Fig. 6. 5-d rainfall at 50% and 80% probabilities.

1,400

1,200

1,000

800

600

400

200

0

Seasonal rainfall (mm)

Total 5-d drought periods

0 2 4 6 8 10 12 14 16 18 20

Y = 1,083 – 45.17X(r = 0.56; significant at 1% level)

1,400

1,200

1,000

800

600

400

200

0

Seasonal rainfall (mm)

Total consecutive 5-d drought periods0 2 4 6 8 10 12 14 16 18 20

50

40

30

20

10

0

Rainfall (mm)

5-d periods by month5 May 5 Jun 5 Jul 5 Aug 5 Sep 5 Oct 5 Nov

50%

80%

800700600500400300200100

0

Cumulative rainfall (mm)

5-d periods by monthJun Jul Aug Sep

50%

80%


15 August), which may happen twice in ten years, the RWS value drops to 0.51. Insuch cases, only about 50% of the crop water requirement would be available fromrainfall and crop yield would be seriously affected.

Yield-rainfall relationshipThe average yields obtained in 1994 and 1995 were 3.6 and 3.5 t ha–1 and the corre-sponding crop field duration rainfalls were 672 and 790 mm, respectively. Both theyield and rainfall in 1994 were below the average for the study area by 0.1 t ha–1 and73 mm, respectively. In 1994, the October rainfall of 204 mm was exceptionally high(probability of 15%) and probably helped keep yield close to the average. Droughtduring the grain-filling stage in 1995 was mainly responsible for the decrease in yield.The total rainfall during October was only 10 mm (probability of 5%). However, 195mm of rainfall during the last five days of September meant that all the fields hadstanding water until the middle of October and water stress developed only in thesecond half of the month. Thus, serious drought did not occur in October and yieldwas not seriously affected.

Total rainfall during crop field duration and the corresponding yields for 1979-93 are plotted in Figure 8. The poor correlation between the two is not unexpected asthe total rainfall during crop field duration can be high, but droughts can still occurduring the critical stages of the crop. Also, changes in yield depend not only on waterbut also on other inputs. Changes in input use are expected because of changes anduncertainties in water availability from year to year. Moreover, the farmers also changedvarieties every 3 to 4 years, and the most common reason cited was to increase ormaintain yields.

Droughts in the study area are expected during October. Hence, the yields (Y)were plotted against the corresponding October rains (R0) to obtain a nonlinear re-gression of the form (Fig. 9)

Fig. 8. Scatter diagram of yield and total sea-sonal rainfall.

Fig. 9. Scatter diagram of yield and Octoberrainfall.

5

4

3

2

1

0

Yield (t ha–1)

October rainfall (mm)

50 100 150 200 250 300

Y = 1.6774 + 0.326 R00.4

(r2 = 0.47; significant at 1% level)

0

5

4

3

2

1

0

Yield (t ha–1)

0 200 400 600 800 1,0001,2001,400 1,600July-October rainfall (mm)

242 Saleh et al

Y = 1.674 + 0.326 R00.4 (6)

with a coefficient of determination (r2) of 0.47 and significant at the 1% level. Thissignifies the importance of rainfall in October for rice yield in the study area.

Drought alleviation: potentials of dry-seeded rice

From the drought analysis, it can be concluded that the study area is drought-prone attwo stages: at the beginning of the season, which can cause a delay in transplanting,and at the beginning of the grain-filling stage, which can drastically reduce crop yield.It has already been mentioned that DSR can be established early and can be veryeffective in drought alleviation. With DSR, drought at the end of the season is notexpected to affect yield because the earlier established crop would be near the har-vesting stage by the time drought set in in October. Moreover, earlier harvested dry-seeded rice would leave a favorable soil-water regime for the subsequent nonricecrop.

A 3-y experimental study (1994-96) on comparative productivity of DSR andTPR at Rajabari by Mazid et al (1998) showed that DSR yields were similar to orslightly better than those of TPR for all seeding/transplanting dates (Table 3). Theexperiments also showed that DSR matured about 1–2 wk earlier than TPR. Becauseof the better soil-water regime, the yield of chickpea planted after DSR was higherthan that after TPR.

Studies in the Philippines have shown that only about 150 mm of rainfall isrequired for crop establishment through DSR compared with about 600 mm for TPR(Saleh and Bhuiyan 1995). Rainfall probability analysis has shown that, if 150 mm ofrainfall is required for establishment of DSR (as in the Philippines with a similar soil

Table 3. Effect of time of seeding (dry-seeded rice, DSR) and transplanting (trans-planted rice, TPR) on grain yield (t ha–1) of high-yielding aman rice at Rajabari,Rajshahi, 1994-96.

Method of Time of seeding/transplantingestablishment

1 June 16 June 1 July 16 July 1 August 15 August

1994DSR – 3.11 3.50 3.45 2.78 –TPR – – – – 2.44 2.77

1995DSR 2.60 2.75 2.85 2.46 – –TPR – – 2.52 2.45 2.93 2.37

1996DSR 3.43 3.93 – – – –TPR – – 3.40 3.81 – –


texture), then once in two years (50% probability) crop establishment may be pos-sible by the first week of June. Twice in ten years (80% probability), however, rainfallmay be inadequate for crop establishment before the third week of June. Present DSRyield data (Table 3) show that even such a delay is not expected to have any effect oncrop yield.

Conclusions

Droughts commonly occur in the study area at the beginning of the season prior totransplanting and from the beginning of the grain-filling stage. In an average year,about 80% of the crop water requirement is available from rainfall. In two out of tenyears, rainfall supplies only about 50% of the total water requirement because of latetransplanting due to early drought. The probabilities of at least one 10-d and 15-ddrought occurring during the grain-filling period are 73% and 53%, respectively. Al-though the total seasonal rainfall is not related to crop yield, there is a fair correlationbetween rainfall in the grain-filling stage and crop yield. Experimental studies ondrought alleviation through dry seeding have been encouraging and the productivityof the dry-seeded rice–chickpea cropping pattern has been higher than that of thetraditional practice. Further research on biophysical and socioeconomic constraintsthat inhibit the wider adoption of dry-seeded rice for drought alleviation as a substi-tute for traditional transplanted rice is needed.

ReferencesGarrity DP, Oldeman LR, Morris RA, Lenka D. 1986. Rainfed lowland rice ecosystems: char-

acterization and distribution. In: Progress in rainfed lowland rice. Manila (Philippines):International Rice Research Institute. p 3-23.

IRRI (International Rice Research Institute). 1993. World rice statistics. Manila (Philippines):IRRI.

Lantican MA, Lampayan RM, Bhuiyan SI, Yadav MK. 1999. Determinants of improving pro-ductivity of dry-seeded rice in rainfed lowlands. Exp. Agric. 35:127-140.

Manalo EB. 1977. Agro-climatic survey of Bangladesh. Bangladesh Rice Research Instituteand International Rice Research Institute. 360 p.

Mazid MA, Mollah MIU, Mannam MA, Elahi NE, Wade LJ. 1998. Increasing productivitythrough rainfed rice-chick pea cropping system in high Barind tract of Bangladesh. Pa-per presented at RLRRC Planning and Review Meeting, Bangladesh Rice Research In-stitute, Gazipur, Bangladesh.

Moya TB, de la Vina WC, Bhuiyan SI. 1986. The potential of on-farm reservoir use in increas-ing productivity in rainfed areas. Philipp. J. Crop. Sci. 2(2):125-132.

MPO (Master Plan Organization). 1985. Crop production limitations in Bangladesh. TechnicalReport No.1, Ministry of Irrigation, Water Development, and Flood Control, Govern-ment of Bangladesh.

Paul DK, Tiwari KN. 1992. Agricultural drought analysis for Hazaribagh, Eastern India. Int.Rice Res. Newsl. 17(6):32-33.

244 Saleh et al

Paul DK, Tiwari KN. 1994. Rainwater storage systems for rainfed ricelands of eastern India:results from research in Hazaribagh District. In: Bhuiyan SI, editor. On-farm reservoirsystems for rainfed ricelands. Manila (Philippines): International Rice Research Insti-tute. p 127-139.

Saleh AFM, Bhuiyan SI. 1995. Crop and rain water management strategies for increasing pro-ductivity of rainfed lowland rice. Agric. Syst. 48(3):259-276.

Steenhuis TS, Van der Molle WH. 1986. The Thornthwaite-Mather procedure as a simple engi-neering method to predict recharge. J. Hydrol. 84:221-229.

Thom HCS. 1968. Direct and inverse tables of gamma distribution. Environmental Data Ser-vice, U.S. Department of Commerce.

Tomar VS, O’Toole JC. 1980. Water use in lowland rice cultivation in Asia: a review of evapo-transpiration. Agric. Water Manage. 3:83-106.

NotesAuthors’ addresses: A.F.M. Saleh, Professor, Institute of Flood Control and Drainage Research,

Bangladesh University of Engineering and Technology, Dhaka 1000, Bangladesh; M.A.Mazid, Principal Agronomist and Head, Rajshahi Regional Station, Bangladesh RiceResearch Institute, Rajshahi 6212, Bangladesh; S.I. Bhuiyan, IRRI Liaison Scientist,GPO Box 64, Ramna, Dhaka 1000, Bangladesh.


Characterizing biotic stresses

246 Jahn et al

Characterizing biotic constraints to production . . . 247

Because rainfed lowland rice (RLR) accounts for 86% of the rice-growing areaof Cambodia, it is important to investigate the biotic constraints to rice inthis ecosystem. Over 3 years, a total of 73 RLR fields in Cambodia werestudied to determine the effect of pests on destabilizing Cambodian RLRyields, which pests and pest combinations affect RLR yields, and how crop-ping practices affect pest levels. Pests included insects, weeds, and dis-eases of rice. Pesticides were not used in any of the fields in this study.Information on crop characteristics and biotic constraints was gathered atfour crop development stages: tillering, booting, milk, and maturity. The inci-dence of damage caused by insects and diseases was recorded from 10hills chosen randomly from each field. Weed infestation was recorded as thepercentage weed cover in three 1-m2 areas. The rice yield of each field wasestimated from an average of three randomly selected 10-m2 areas. Corre-spondence analysis was used to characterize the patterns of cropping prac-tices, pest infestations, environmental conditions, and yields. The results ofthis analysis generated testable hypotheses about the factors contributingto pest problems. Fields of early duration rice tended to have low levels ofhispa and high levels of Pentatomids, while late-duration fields had high hispaand low Pentatomid levels. Fields without standing water had higher thanaverage levels of weeds, cutworm, hispa, and Pentatomids. Brown spot andnarrow brown spot were the only diseases observed frequently enough tomake inferences about their relation to cropping practices.

As a component of systems research in rice plant protection, this studyassisted in predicting the effects of cropping practices on pest infestations.These techniques have several limitations, however: the danger of drawingfalse conclusions, difficulties in interpreting results, insufficient attention tothe soil type and relative rates of fertilizer, the inability to capture time ad-equately as a variable, the lack of information on the relative contribution ofpests to variation in yield data, the incomplete coverage of pests, and thefallibility of observers. Each limitation is discussed in detail. The results ofthis study contributed to an assessment of relative pest importance andthereby helped prioritize research to develop integrated pest managementrecommendations.

Characterizing biotic constraintsto production of Cambodianrainfed lowland rice: limitationsto statistical techniquesG.C. Jahn, Pheng Sophea, Pol Chanthy, and Khiev Bunnarith

248 Jahn et al

Rainfed lowland rice (RLR) accounts for 86% of the rice-growing area of Cambodia(Javier 1997). Cambodian rice productivity is the lowest in Asia due to the high per-centage of RLR. The reasons for low RLR productivity include erratic rainfall, poorsoils, low seed quality, socioeconomic constraints, and pest damage (Javier 1997).Cambodian rice farmers often cite poor water availability and pests as the major causesof low and unstable yields (Jahn et al 1997). This is understandable since poor soils,low seed purity, and socioeconomic factors (such as lack of farm machinery) areever-present constraints to yield that will not completely destroy a crop. In contrast,droughts, floods, and pest outbreaks can visibly reduce the yields of a farmer’s fieldcompared to previous years.

From 1994 to 1996, a total of 73 RLR fields in Cambodia were studied to deter-mine the effect of pests on RLR yield stability, which pests and pest combinationsaffect yields, and how cropping practices affect pest levels. Systematic survey data onpest losses are an important part of the information needed to quantify the risk of pestinjuries (Cohen et al 1998). The Cambodia-IRRI-Australia Project (CIAP) IntegratedPest Management (IPM) Program uses this information to determine which pests arethe most important, which helps prioritize research. We also use this information topredict the effect of cropping practices on pest infestations and develop IPM recom-mendations. Using databases of categorical information to characterize the patternsof cropping practices, pest infestations, environmental conditions, and yields is animportant component of CIAP’s systems research in rice plant protection (CIAP 1998).


Pest constraints to RLR yields were characterized based on the methods of Savary etal (1995, 1996) developed at the International Rice Research Institute (IRRI). “Pests”included insects, weeds, and diseases. The data collection methods of Savary et al(1996) were revised to reflect the conditions and pests of the Cambodian RLR eco-system. The observations were restricted to certain pests. The selection of pests in-cluded in the study was based on their perceived importance according to farmers,extension workers, and scientists. Farmer interviews (Jahn et al 1997), pest collec-tions, and practical considerations (e.g., ease of recognizing the pest or the damage)were used to determine which pests to include in the study of pest constraints to RLRproduction.

Over 3 years, 1994 to 1996, a total of 73 RLR farmers’ fields (>0.3 ha each)were surveyed in Phnom Penh, Svay Rieng, and Takeo provinces. All surveys wereconducted in the wet season. Pesticides were not used in any of the fields in this study.Information on the crop characteristics and the biotic constraints was gathered at fourcrop development stages: tillering, booting, milk, and maturity. The yield of eachfield was estimated by crop cuts from an average of three randomly selected 2 m × 5m areas at ripening stage, converted to t ha–1 and adjusted to 14% moisture. Theincidence of pests was recorded from 10 hills chosen haphazardly from each field.Weed infestation was assessed on the basis of the percentage weed cover in three1-m2 areas that included sampling hills number 3, 6, and 9. Rat damage was not


included in these studies. Later studies, however, indicated the importance of rats andthey were included in follow-up studies from 1997 to 1999 (CIAP 1998, 1999).

The levels of gall midges and stem borers were measured by the number ofinfested tillers per hill. Damage by whorl maggots, leaffolders, hispa, and cutwormswas measured by the number of damaged leaves per hill. Planthoppers, rice bugs, andother sucking insects were directly counted from 10 hills, whereas leafhoppers andbeneficial insects were counted from 5 sweep-net samples per field per visit. Eachsweep-net sample consisted of 10 sweep-net strokes. Rice bugs were counted directlyrather than assessed on the basis of unfilled or partially filled grains. This type ofdamage could be caused by other pests or even by thermal damage (Satake and Yoshida1978, Sheehy et al 1998). Rice diseases were quantified by counting the number ofdamaged leaves, tillers, or panicles per hill depending on the symptoms of the par-ticular disease.

A team of two to three trained observers took approximately 1 h to make (andrecord) all of the necessary observations on each visit to a farmer’s field.

Data analysisThe analysis proceeded in five steps: (1) determination of average injury levels ofeach pest for each crop stage, (2) categorization of the variables (i.e., pest, croppingpractices, and yield) into classes, (3) testing for independence of paired variables(i.e., pest levels, cropping practices, and yields) in contingency tables, (4) clusteringof cropping practices and pest profiles, and (5) development of contingency tablesand correspondence analysis. Data were analyzed using the computer programs Ex-cel® and STAT-ITCF®.

Average injury levels and selection of peak period. The percent damage in-flicted by each particular type of pest was averaged over the fields at successive cropdevelopment stages and graphed (Figs. 1 to 7) to indicate changes in injury and pestlevels over time. Only data from the peak average injury level for each pest wereanalyzed thereafter (Savary et al 1996).

Transformation of quantitative variables into classes. All levels within a classcontained a similar number of rice fields, i.e., classes were made as balanced as pos-sible. Yields and injury levels were placed into categories of low, medium, and highso that qualitative and quantitative variables could be analyzed simultaneously in

1

0

Weed infestation rating

Tillering Booting Milky RipeningCrop stage

Fig. 1. Average weed infestation in 73 rainfed low-land rice fields. All weeds were below the ricecanopy; ■ = average level of weeds below ricecanopy.

250 Jahn et al

Fig. 2. Average levels of tiller damage by gall midgesand stem borers in 73 rainfed lowland rice fields.Deadhearts and whiteheads are both symptoms ofstem borer damage.

Fig. 3. Average levels of leaf damage by whorl maggot,leaffolder, hispa, cutworm, and other leaf-feeding in-sects in 73 rainfed lowland rice fields.

Fig. 4. Average levels of planthoppers, rice bugs, andother sucking insects in 73 rainfed lowland rice fields.

4

3

2

1

0

Average % tillers affected


Gall midgeDeadheartsWhiteheads

20

15

10

5

0

Average % leaf affected


Whorl maggotLeaffolderHispaCutwormLeaf-feeding insects

Tillering Booting Milky Ripening

Crop stage

2

1

0

Average insect number

PlanthoppersRice bugsOther sucking insects


contingency tables. Tables 1 to 3 summarize the classes, boundaries, and frequencydistribution of the qualitative and quantitative variables.

Tests of independence for relationships among pests, cropping practices, andyields. Tests for independence were performed only on variables with balanced classes.The cells of contingency tables (particular pest × particular cropping practice) werefilled in with the number of fields that matched each pair of classification criteria.Chi-square tests were used to interpret relations between any paired variables. Rela-

Fig. 5. Average number of leafhoppers, pests, andnatural enemies per sweep-net sample of 40 rainfedlowland rice fields.

Fig. 6. Average incidence of foliar diseases at differ-ent crop stages in 73 rainfed lowland rice fields.

Fig. 7. Average percentage of tiller damaged by twofungal diseases at different crop stages in 73 rainfedlowland rice fields.

Average % tiller or panicle affected


1

0

Sheath blightSheath rot

40

30

20

10

0

Average number of insects per sweep


LeafhoppersPestsNatural enemies

40

30

20

10

0

Average % leaf affected


Brown spot

Bacterial leaf streak

Narrow brown spot

252 Jahn et al

Table 1. Categorization of cropping practices into classes.

Variable Label Class boundaries Individuals(no.)

Crop duration CD1 Early 21CD2 Medium 34CD3 Late 18

Fertilization F1 0–30 kg ha–1 of mineral fertilizer (MF) 17F2 >30–150 kg ha–1 of MF 24F3 >150 kg ha–1 of MF 19F4 Manure 13

Yield Y1 Very low yield (0–1.85 t ha–1) 19Y2 Low yield (>1.85–2.51 t ha–1) 17Y3 Medium yield (>2.51–3.31 t ha–1) 19Y4 High yield (>3.31 t ha–1) 18

Water status WST1 No or little water (1–7 cm) 32WST2 Adequate water (>7–8 cm) 19WST3 Too much water (>8 cm) 22

tionships with P <0.05 are summarized in Table 4. Pests without any detectable rela-tion to any cropping practice were omitted from the table.

Categorize cropping practices and pests into clusters. Four clusters of crop-ping practices (CP) and pest profiles (PE) were generated by STAT-ITCF using clus-ter analysis (Tables 5 and 6). Only pest variables related to at least one of the CPclusters were included in the analysis.

Contingency tables and correspondence analysis. Contingency tables were con-structed to show the frequency distribution of RLR fields as a function of yield andcropping practices, and as a function of yield and pest profiles (Table 7). The two (4 ×4) contingency tables (PE × Y and CP × Y) were linked to produce a 4 × (4 + 4) datamatrix for correspondence analysis and graphed on factorial axes (Savary et al 1996)(Table 8, Fig. 8). Two factorial axes were sufficient to display the positions of rowand column points. The display could be completed with three axes, but the interpre-tation is more complex.


Table 2. Categorization of weed and arthropod data into classes.

Variable Class boundaries Individuals (no.)

Weeds (WB) Very low (score 0–0.2) 25Medium (>0.2–0.4) 19High (>0.4–2.0) 29

Gall midge (GM) Low (0–0.2%) 25Medium (>0.2–1.0%) 23High (>1–59.3%) 25

Deadhearts (DH) None (0%) 42Low (>0–0.5%) 14Medium (>0.5–4.1%) 17

Whiteheads (WH) Low (0–0.3%) 24Medium (>0.3–1.0%) 26High (>1–4.9%) 23

Whorl maggot (WM) Low (0–1.1%) 25Medium (>1.1–2.9%) 24High (>2.9–14%) 24

Leaffolder (LF) Low (0–0.2%) 25Medium (>0.2–0.7%) 23High (>0.7–4.2%) 25

Hispa (HP) Low (0–0.9%) 25Medium (>0.9–4.8%) 24High (>4.8–12.8%) 24

Cutworm (CW) Low (0.8–8.1%) 25Medium (>8.1–15.4%) 24High (>15.4–51.1%) 24

Leaf-feeding insects Low (0–0.1%) 25(LFI) Medium (>0.1–2.5%) 23

High (>2.5–32%) 25Planthoppers (PH) Low (0–0.1 PH hill–1) 33

Medium (>0.1–0.4) 21High (>0.4–14.8) 19

Rice bug None (0 RB hill–1) 56Low (>0–0.1) 9Medium (>0.1–0.6) 8

Other sucking insects Low (0–0.1 SI hill–1) 25(SI) Medium (>0.1–0.9) 24

High (>0.9–16.3) 24Leafhoppers (LH) Low (3.1–14.4 LH sweep net–1) 14

Medium (>14.4–31.1) 13High (>31.1–104.5) 13

Pest (PS) Low (3.1–8.8 PS sweep net–1) 14Medium (>8.8–16.8) 13High (>16.8–65.7) 13

Natural enemies (NE) Low (6.2–15 NE sweep net–1) 14Medium (>15–19.8) 13High (>19.8–51.3) 13

254 Jahn et al

Table 4. Relationship of management practices to pests in rainfed lowland rice.a

Pentatomids NarrowManagement (stink bugs Whorl Brown brownpractice Weeds Cutworm Hispa and maggot spot spot

black bugs)

Cultivar durationLate – ? + – N N NMedium N N N N N N NEarly + N – + N – N

Water levelNo water + + + + N ? ?1–5 cm N N – – N ? ?>5 cm – – N N N ? ?

Chemical fertilizer0–30 kg ha–1 ? – N N N ? N31–150 kg ha–1 ? N N N N + N>150 kg ha–1 ? + ? N + – –

a + = increases pest population, – = decreases pest population, N = no effect, ? = unknown, i.e., more datarequired.

Table 3. Categorization of disease data into classes.

Variable Class boundaries Individuals (no.)

Brown spot (BS) Low (0–12.3%) 25Medium (>12.3–32.1%) 24High (32.1–93.7%) 24

Bacterial leaf streak (BLS) None (0%) 44Low (>0–5.5%) 15High (>5.5–30%) 14

Narrow brown spot (NBS) None (0%) 27Low (>0–4.2%) 23High (>4.2–35.9%) 23

Sheath blight (SHB) None (0%) 61Low (>0–0.6%) 6Medium (>0.6–2.7%) 6

Sheath rot (SHR) None (0%) 51Low (>0–0.5%) 11Medium (>0.5–4.1%) 11


Table 5. Properties of individual clusters of cropping practices (CP).

Clusters Crop duration Fertilizer Water status

CP1 (25 fields) Late varieties and some No to low amount of Little watermedium varieties mineral fertilizer

CP2 (25 fields) Medium varieties and High mineral fertilizer Good watersome early varieties application and manure management

CP3 (11 fields) Medium rice varieties Medium to high levels of Too much watermineral application

CP4 (12 fields) Early varieties No to low amount of Variablemineral fertilizer

Table 6. Properties of the pest profile (PE) clusters.

Pest profile Characterized by high levels of these pests

PE1 (29 fields) Weeds, whorl maggots, hispa, narrow brown spot,and sucking insects

PE2 (22 fields) Narrow brown spotPE3 (13 fields) Brown spot, gall midges, and sucking insectsPE4 (9 fields) Gall midge, cutworm, brown spot, and weeds

Table 7. Contingency table showing the numberof fields matching each yield (Y) profile and eachcluster of cropping practices (CP) and pest pro-files (PE).

Yield profilesCluster

Y1 Y2 Y3 Y4

CP1 13 5 3 4CP2 1 5 9 10CP3 3 2 4 2CP4 2 5 3 2PE1 9 5 9 6PE2 1 7 6 8PE3 4 3 2 4PE4 5 2 2 0

256 Jahn et al

Table 8. Numerical output of the correspondence analysis of Table 7.

Eigen values (varianceson principal axes) Contribution to total variances Cumulated percentage

0.1844 79.7% 79.7%0.0269 11.6% 91.3%

Principal axis 1 Principal axis 2

Columns Squared Relative Squared RelativeCoordinates cosine contribution Coordinates cosine contribution

(%) to axis 1 (%) to axis 2

Y1 +0.702 0.992 69.5 –0.041 0.003 1.6Y2 –0.102 0.109 1.3 +0.280 0.824 67.9Y3 –0.227 0.467 7.3 –0.174 0.272 29.1Y4 –0.405 0.789 21.9 –0.038 0.007 1.3

Principal axis 1 Principal axis 2

Rows Squared Relative Squared RelativeCoordinates cosine contribution Coordinates cosine contribution

(%) to axis 1 (%) to axis 2

CP1 +0.588 0.935 32.1 +0.048 0.006 1.4CP2 –0.550 0.930 28.0 –0.143 0.063 12.9CP3 +0.039 0.021 0.1 –0.184 0.487 9.5CP4 –0.116 0.065 0.6 +0.367 0.656 41.0PE1 +0.107 0.277 1.2 –0.159 0.615 18.7PE2 –0.488 0.900 19.5 +0.159 0.095 14.1PE3 +0.077 0.088 0.3 +0.083 0.102 2.3PE4 +0.738 0.926 18.2 +0.006 0.000 0.0

Fig. 8. Graphical display provided by correspondence analysis of the data matrix in Tables 7 and8. The cropping practice (CP), yield (Y), and pest (PE) clusters can be grouped into four do-mains described in Table 9.

CP4

PE2

Y2

Y4

CP2Y3 CP3

PE1

PE3PE4

Y1

Axis 1

Axis

2

0.8–0.8

0.4

–0.4

1.6

00

CP1

Simultaneous representation of rows (observations)and columns (variables)

A

B

C

D


Results

Relations among pests and cropping practicesLate-duration rice fields tended to have high levels of hispa and low levels ofPentatomids, whereas early duration fields were the opposite. Fields without standingwater at the four key crop stages had higher than average levels of weeds, cutworm,hispa, and Pentatomids. Brown spot and narrow brown spot were the only diseasesobserved frequently enough in the RLR production systems to make inferences abouttheir relation to cropping practices. Other diseases, such as sheath blight, are associ-ated with intensive and high-input production systems (Cu et al 1996). Fields receiv-ing more than 150 kg fertilizer ha–1 had lower than average levels of brown spot andnarrow brown spot (Table 4). The fertilizer associations would depend, of course, onthe type of fertilizer used, not only the rates. Later experiments addressed the associa-tion between type of fertilizer and pest levels (CIAP 1998, 1999).

Correspondence analysisThe first and second axes of the graphical display of the correspondence analysis(Fig. 8) accounted for 79.7% and 11.6% of total inertia, respectively. Therefore, thefirst two axes provided a good overall view of the numerical output of the correspon-dence analysis (Table 8), as 91.3% of the total inertia was represented. Axis 1 repre-sents the gradient of decreasing levels of rice yields. It involves the contributions ofY1, Y4, CP1, CP2, PE2, and PE4 (Tables 5, 6, and 7). Axis 2 involves the contribu-tions of Y2, Y3, CP3, CP4, PE1, and PE3. Four domains were generated from Figure8—domain A (Y1, CP1, PE4), domain B (Y2, CP4, PE3), domain C (Y3, CP3, PE1),and domain D (Y4, CP2, PE2)—by correspondence analysis using STAT-ITCF (Savaryet al 1996). Table 9 summarizes the characteristics of each domain.

Correspondence analysis of Table 7 indicates that CP1 (Table 5) tends to havelower yields, whereas CP2 tends to have medium to higher yields. Fields with a pestprofile matching PE2 (Table 6) are more frequently associated with high yields, butfields matching PE4 more frequently have low yields (Table 8, Fig. 8).

Discussion

The effects of cropping practices on the levels of pest infestation (Table 4) are basedon chi-square tests of independence. While this type of analysis indicates whether aparticular practice (or condition) tends to be associated with high or low levels ofparticular pests, it does not measure the relative strength of that association (or possi-bly effect). The interactions between combinations of cropping practices, multiplepests, and yields (Fig. 8, Table 9) are based on correspondence analysis. In this case,the relative strength of these relationships is indicated by the distance between clus-ters in Figure 8.

258 Jahn et al

Examples of predictions based on findingsNumerous predictions were generated by the rice pest constraint study. Table 4 sum-marizes some of these predictions; other examples include:

● Fields of medium-duration varieties with high mineral fertilizer (>150 kgha–1) and manure applications and good water management have lower thanaverage levels of pests in general, except narrow brown spot disease. Thesefields tend to have yields higher than 3.3 t ha–1.

● Fields of late-maturing varieties with 0 to 30 kg ha–1 of mineral fertilizer andinsufficient water have higher than average levels of weeds, cutworm, gallmidge, and brown spot. These fields tend to have yields less than 1.9 t ha–1

(Table 9).● Leaf-feeding insects (taken as a single group) do not constrain yields under

any combination of cropping practices when less than 33% of the leaves aredamaged.

● Tall, late-maturing rice varieties compete better against weeds than short,early maturing rice varieties in the RLR ecosystem.

Assessing the accuracy of predictionsThe results of these analyses were converted into hypotheses that were tested in on-station or on-farm experiments as part of the systems approach to rice plant protec-tion research (Fig. 9). For example, the relationship between rice cultivar duration

Table 9. Characteristics of the domains derived from Figure 8.

Domain Clusters Cropping practices and yields Pest constraints

A Y1, CP1, PE4 Late-duration rice varieties Weeds, gall midge, cutworm,and some medium varieties and brown spot

Low application of mineralfertilizer

Diverse water depthsVery low yield

B Y2, CP4, PE3 Early rice varieties Brown spot, gall midge,Low application of mineral sucking insects

fertilizerDiverse water depthsLow yield

C Y3, CP3, PE1 Medium rice varieties Weeds, whorl maggot,High application of mineral hispa, sucking insects,

fertilizer and gall midgeToo much waterMedium yield

D Y4, CP2, PE2 Medium cultivars and some Narrow brown spotearly varieties

High application of mineralfertilizer and manure

Adequate waterHigh yield


Fig. 9. Schematic diagram of the systems research in rice plant protection developed and usedby the IPM Program of the Cambodia-IRRI-Australia Project. IPM = integrated pest manage-ment.

Interview farmers todocument their

knowledge, practices,and perceptions about

rice pests

Collect and identify floraand fauna of the rice

ecosystems

Characterization of bioticconstraints to yield

Generation of testablehypotheses

Conduct experiments totest hypothetical

relationship betweencropping practices, pests,

and yields

Project likely future pestproblems in relation tochanges in agriculture,

e.g., new cultivars,fertilizers

Gather pest distributiondata from collections,

farmers, extensionservices, news reports,

etc.

Verify reports

Build database of pestdistribution in time

and space

Create pest distribution maps

Forecast likely pest problems

● Farmer field schools● Extension● Farmer participatory research● Demonstrations● Media● Farmer experiments

Experiments to evaluateeffectiveness of

farmers’ practices

Experiments to improveon farmers’ pest

management

Test integration of pestprevention and controltechniques with farmers

IPM options andrecommendations

Adoption and adaptation of IPM recommendations by farmers

Desired outcomes:● Stable yields● Increased average yields● Less costly pest management● Safer pest management● Sustainable pest management

Goals:● Increased net household

income● Improved food security

Start here

260 Jahn et al

and weeds (Table 4) suggests that late rice varieties will compete against weeds betterthan early rice varieties in the RLR ecosystem. An experiment testing this hypothesisled to the conclusion that CAR8, a late-duration rice, is a better competitor againstweeds than IR66, an early duration rice, in the RLR ecosystem (CIAP 1998). In an-other experiment, cutting off different amounts of rice leaves at different stages indi-cated that, before the booting stage, leaf-eating insects do not constrain rice yields(Khiev et al 1999), supporting the conclusions of the pest constraint study.

The results of the correspondence analysis (Fig. 8, Table 9) suggest that theproper combination of rice cultivar, fertilizer application, and water management canminimize pest problems and thereby increase and stabilize yields (by removing thelower extremes of the yield variation). To test this hypothesis, we measured pest lev-els and yields in farmers’ fields treated with a factorial combination of two rice vari-eties and three fertilizer rates, for a total of six plots per field. We collected data from157 fields in 1997 and from 104 fields in 1998 (CIAP 1998, 1999). Multiple regres-sion analysis of the resulting database reveals which pests contribute to variations inyield under different combinations of soil types, fertilizers, rice cultivars, and waterlevels. Some general conclusions that apply to all RLR in Cambodia can also beelucidated from this database. For example, tillering-stage insect pests had little im-pact on yields in RLR, whereas rats were an important source of yield loss (CIAP1998, 1999).

Assessing pest importance and prioritizing researchWe assess the importance of pests based on farmers’ perceptions, the impact of thepest on yield, the farmers’ ability to manage the pest, and the estimated risk of dam-age to the crop (Table 10). These factors are used to prioritize IPM research, takinginto account whether the farmers’ pest management practices are sustainable (Fig.10).

Limitations of the techniquesWhile the pest and injury survey techniques described in this chapter are useful forpredicting the effects of cropping practices on pest infestations, these techniques haveseveral limitations: the danger of drawing false conclusions, difficulties in interpret-ing results, insufficient attention to the soil type and relative rates of fertilizer, theinability to capture time adequately as a variable, the lack of information on the rela-tive contribution of pests to variation in yield data, the incomplete coverage of pests,and the fallibility of observers. Each of these limitations is discussed in detail below.These limitations do not render the statistical analysis useless, but they must be con-sidered when interpreting the results, deriving hypotheses, and planning further re-search based on the results.

It is important to note that the characterization of biotic constraints to rice pro-duction through independence testing and cluster analysis is a means of generatingtestable hypotheses rather than an end in itself. Although these techniques are power-ful for ruling out factors as major constraints to rice production, they only revealassociations and correlations of factors rather than cause and effect. Researchers must


Tabl

e 1

0.

Info

rmat

ion

used

to

prio

riti

ze r

esea

rch

and

asse

ss o

ppor

tuni

ties

for

gai

ns.

Pote

ntia

l yie

ldPr

obab

ility

of

Farm

erD

o fa

rmer

s’R

esis

tant

or

Pro

nelo

ss in

indi

vi-

yiel

d lo

ss in

asse

ssm

ent

Farm

ers

atte

mpt

s to

com

petit

ive

Pest

sS

tatu

sato

out

brea

ks

dual

see

dbed

an

indi

vidu

alof

pro

blem

inkn

ow h

ow t

om

anag

e it

lead

rice

varie

ties

in C

ambo

dia

or fie

ldfie

ld if

the

pes

ten

dem

ic a

rea

man

age

itto

uns

usta

inab

leav

aila

ble?

is n

ot m

anag

edpr

actic

es?

Inse

cts

Bro

wn

plan

thop

per

AYe

sH

igh

Low

Maj

orN

oYe

sYe

sS

tem

bor

erC

No

Med

ium

Low

Maj

orN

oYe

sYe

sG

all m

idge

CN

oM

ediu

mM

ediu

mM

ajor

No

Yes

Yes

Gra

ssho

pper

CYe

sH

igh

Low

Maj

orYe

sYe

sN

oR

ice

bug

CN

oM

ediu

mH

igh

Maj

orYe

sYe

sN

oAr

myw

orm

AYe

sH

igh

Low

Maj

orN

oYe

sN

oTh

rips

AYe

sM

ediu

mM

ediu

mM

ajor

No

Yes

No

Cas

ewor

mC

No

Low

Low

Maj

orN

oYe

sN

oLe

affo

lder

CN

oLo

wLo

wM

ajor

No

Yes

No

Cric

ket

CN

oLo

wLo

wM

ajor

Yes

Yes

No

Wee

dsLy

thra

ceae

CN

oH

igh

Hig

hM

inor

Yes

No

No

Cyp

erus

dif

form

isC

No

Hig

hH

igh

Min

orYe

sN

oN

oC

. iri

aC

No

Hig

hH

igh

Min

orYe

sN

oN

oEc

hino

chlo

a co

lona

CN

oH

igh

Hig

hM

inor

Yes

No

Yes?

(alle

lopa

thic

ric

e)E.

cru

s-ga

lliC

No

Hig

hH

igh

Min

orYe

sN

oYe

s? (co

mpe

titiv

ecu

ltiva

rs)

Dis

ease

sTu

ngro

AN

oLo

wLo

w?

No

No

Yes

Bro

wn

spot

CN

oLo

wLo

w t

o m

ediu

mM

inor

Yes

No

No

Bla

stA

Yes

Med

ium

Low

Min

orN

oN

oN

oS

heat

h ro

tC

No

Low

Low

Min

orN

oN

oN

oO

ther

sR

ats

C &

AYe

sH

igh

Hig

hM

ajor

Yes

Yes

No

Cra

bsC

& A

No

Hig

hH

igh

Maj

orYe

sYe

sN

o

a A =

acu

te, C

= c

hron

ic.

262 Jahn et al

Fig. 10. Process for prioritizing research to develop sustainable pest managementpractices.

proceed with caution when interpreting the results of such analyses. An illustration ofthis is provided by the data on cutworm. Higher than average levels of cutwormswere associated with an absence of water in rice fields (Table 4). Naturally, fieldswithout water have low yields, so it is not surprising that the cluster analysis (Fig. 8)places the lowest yielding fields and cutworm-infested fields in the same domain(Table 9). To conclude from Table 9 that cutworms are a major constraint to RLRproduction, however, would be rash.

Do farmers considerthis a major pest?

No Yes

Does itconstrain riceproduction?

No Yes No Yes

May need to reevaluatefarmers’ views and effect

of pest on yield

Does itconstrain riceproduction?

Determinewhy farmers

hold this viewAre farmers’

attempts to manage itunsustainable?

No Yes

Address farmers’misconceptions through

farmer field schools,extension, etc.

Address scientists’misconceptions

through publications

No furtherresearch

Low researchpriority

High researchpriority


Interpreting the results of independence tests and cluster analysis also presentsdifficulties. While independence tests indicate that late-maturing varieties discourageweeds better than early maturing varieties (Table 4), correspondence analysis resultedin the formation of domain A (Table 9), which contains traditional varieties, low yields,and higher than average weed levels. In other words, fields with traditional varietiesand high weed levels had lower yields, but this does not mean that traditional ricevarieties encourage weeds. On the contrary, laboratory screening has revealed that atleast 11 traditional Cambodian rice cultivars exhibit allelopathic properties againstawnless barnyard grass, Echinochloa colona (L.) Link, while the IR varieties testeddo not exhibit such traits (Pheng et al 1999).

This assessment of pest constraints to RLR production did not include data onthe soil types, nor did we analyze the data in terms of relative amounts of NPK orother minerals applied. The effect of soil nutrient stress on pest damage may be indi-rect, but profound. For example, nutrient stresses may delay crop development (Kirket al 1998), resulting in crops that are out of synchrony with crops in the surroundingarea, and therefore at greater risk of pest damage. Detailed soil and fertilizer variableswere recorded and taken into account in the follow-up study designed for multipleregression analysis, described above (CIAP 1998, 1999).

Another limitation of this technique arises from the assessments of injuries orpest levels over four successive development stages. There are three ways to dealwith the large data set of multiple pests spread over time. First, the entire data setcould be analyzed: carrying out the analysis for each pest for each stage. Not onlywould this process be time-consuming, but the results would be difficult to interpret,since different pests affect the crop at different stages. For instance, an analysis oftillering-stage pests may reveal the relationship of gall midge and stem borer to yield,but not rice bugs, which are primarily pests at the milk stage. Analyzing all of the databy stage would not reveal whether rice bugs or gall midges represent a greater con-straint to RLR production.

A second method would be to compact the data over time: creating a new vari-able based on the area under a curve or an average of the pest/injury levels at eachcrop stage (Jahn 1992, Savary et al 1996, Teng and Bissonnette 1985). Taking thearea under the curve would be quite misleading for some of the pest variables, how-ever. Some types of damage are retained from one stage to the next. Tillers damagedby gall midge at the tillering stage will still be visible at the booting stage, but notvisible by the milk stage (Fig. 2). Taking the area under the gall midge curve wouldlead to an exaggeration of the amount of gall midge damage, since most of the gallsrecorded at tillering would be recorded again at booting. Other types of injury, suchas whitehead resulting from stem borer damage, only appear at a certain stage. Thearea under the curve for whitehead (Fig. 2) would include the large area created byconnecting the data point, i.e., zero, at booting to the data point at the milk stage,when in fact no whitehead is visible before the milk stage. An additional problem isthat the area under the curve would depend on the graphed distance from one stage toanother. Since the study included numerous rice varieties, the number of days be-tween stages varies greatly. Attempting to incorporate that information into the analy-

264 Jahn et al

sis would make it unwieldy. While taking an average of the pest/injury levels acrossstages would avoid the problem of gauging the time between crop stages and analyz-ing areas under the curve, the problem of counting some damage twice (or more)would remain.

A third approach to the problem would be to analyze data from the peak periodof average injury. This is the approach that we chose. Like the creation of a newvariable, analyzing peaks results in a considerable reduction in the number of vari-ables to examine, but with the added advantage of avoiding the issues of the timebetween crop stages, areas under curves, and recounted observations. Analyzing peakdata has the additional advantage that the pest/injury variables are compared witheach other when they cause the greatest harm and are the easiest to observe. The chiefdisadvantage of this approach is that it cannot distinguish between a field with highlevels of a pest at a single stage and another field with similarly high levels of thesame pest but over several stages. A field that is continuously defoliated from tilleringto the milk stage might be expected to have a lower yield than a field that is onlydefoliated at the booting stage, even if the two fields have similar levels of defoliationat the booting stage. Keeping this limitation in mind, we were careful not to makeinferences about the relationship between specific pests and yields when we weretesting for independence of variables (Table 4).

Cluster analysis does not reveal the relative contribution of variables (e.g., pestspecies, rice cultivar, fertilizer rates) to the variation in yield data. Multiple regres-sion analysis might provide this information; however, this experiment was not de-signed for multiple regression analysis. Some of the variables in the database werealways associated with each other (e.g., certain rice cultivars were found only at onelocation, and all without fertilizer), making it impossible to distinguish their relativecontributions to variations in yield. The factorial combination experiments conductedin 1997 and 1998 (described above) were designed for analysis by multiple regres-sion and corrected for this limitation (CIAP 1998, 1999).

Finally, the selection of key pests to include in the survey presents several limi-tations. First, this type of study characterizes pest constraints to RLR production basedon the fields already growing rice. Pests (e.g., rats) that reach such an intensity thatthey discourage farmers from growing rice at all could not be included in this survey.Instead, we gather information (on pests that prevent crops from being grown) byinterviewing farmers (Jahn et al 1997). Although water may be adequate to grow anearly duration variety in the early wet season, farmers are generally loath to do sobecause of the extreme pest damage that these crops suffer. A second problem is thatonly certain key pests are recorded in the study, which automatically excludes rarebut potentially important pests from the database. The CIAP IPM Program, however,has ongoing studies of RLR arthropod community structure and biodiversity as partof an attempt to classify all of the flora and fauna of the Cambodian RLR ecosystem(CIAP 1997, 1998, 1999). Recently introduced pests, such as the golden apple snailand the rice leaf weevil, have been discovered and monitored as a result of thesebiological inventories (Jahn et al 1998, CIAP 1996). A third problem is that pestdamage often does not permit identification of the pest to the species, or even genus,


level. Again, we augment our database through separate field collections. For ex-ample, rat damage is easily recognized, but only by trapping specimens have we beenable to determine which rat species are attacking RLR in Cambodia (Jahn et al 1999).Fourth, the quality of the data must also be considered. It is often difficult to distin-guish different types of pest damage. All of the cooperators joining us in data collec-tion have undergone extensive training in identifying, quantifying, and recording pests/injuries of RLR. Still, there are bound to be mistakes when dealing with such a largedatabase. All the findings and predictions must be considered hypotheses that aresubject to more rigorous testing.

Conclusions

Nonparametric statistical techniques (Savary et al 1996) can be valuable tools forgenerating testable hypotheses (but not conclusions) on how cropping practices arerelated to pest levels, and how those pest levels are related to yield constraints. Non-parametric analysis not only reveals associations and correlations but also helps ruleout factors as major constraints to rice production over a large area. For example,leaf-feeding insects did not constrain yields under any combination of cropping prac-tices if less than 33% of the leaves were damaged at the booting stage.

The results of such analysis, however, are only a partial contribution to under-standing biotic yield constraints. Relations among pests and cropping practices can-not be adequately described or even quantified simply based on statistical techniquesthat do not reveal causality. Nor can these nonparametric statistical techniques deter-mine the impact of pests on yields at the field or farm level, where research resultsmust ultimately be applied. Statistical descriptions of pest constraints should be inter-preted in the context of the biology and ecology of the pest, the physical environment(e.g., soil type), economic and social conditions, the experience of rice farmers, andthe manner in which farmers react to perceived pest problems. The value of generat-ing testable hypotheses by statistical techniques must be evaluated by the cost ofgathering such data, the reliability of such data, and the ultimate application of theanalysis.

ReferencesCIAP (Cambodia-IRRI-Australia Project). 1996. Annual research report 1995. Phnom Penh

(Cambodia): CIAP. 185 p.CIAP (Cambodia-IRRI-Australia Project). 1997. Annual research report 1996. Phnom Penh



(Cambodia): CIAP. 205 p.

266 Jahn et al

Cohen MB, Savary S, Huang N, Azzam O, Datta SK. 1998. Importance of rice pests and chal-lenges to their management. In: Dowling NG, Greenfield SM, Fischer KS, editors.Sustainability of rice in the global food system. Davis, Calif. (USA): Pacific Basin StudyCenter, and Manila (Philippines): International Rice Research Institute. p 145-164.

Cu RM, Mew TW, Cassman KG, Teng PS. 1996. Effect of sheath blight on yield in tropical,intensive rice production system. Plant Dis. 80(10):1103-1108.

Jahn GC. 1992. Effect of neem oil, monocrotophos, and carbosulfan on green leafhoppers,Nephotettix virescens (Distant) (Homoptera: Cicadellidae) and rice yields in Thailand.Proc. Hawaiian Entomol. Soc. 31:125-131.

Jahn GC, Cox P, Mak Solieng, Chhorn Nel, Tuy Samram. 1999. Rat management in Cambodia.In: Singleton G, Hinds L, Leirs H, Zhibin Zhang, editors. Ecologically-based rodentmanagement. Canberra (Australia): Australian Centre for International Agricultural Re-search.

Jahn GC, Pheng S, Khiev B, Pol C. 1997. Pest management practices of lowland rice farmersin Cambodia. In: Heong KL, Escalada MM, editors. Pest management practices of ricefarmers in Asia. Manila (Philippines): International Rice Research Institute. p 35-51.

Jahn GC, Pheng S, Khiev B, Pol C. 1998. Pest potential of the golden apple snail in Cambodia.Cambodian J. Agric. 1:34-35.

Javier E. 1997. Rice ecosystems and varieties. In: Nesbitt HJ, editor. Rice production in Cam-bodia. Manila (Philippines): International Rice Research Institute. p 39-81.

Khiev B, Jahn GC, Pol C, Chhorn N. 1999. Simulating rice pest damage to determine effectson yield. Cambodian J. Agric. 2(1):29-32.

Kirk GJD, Dobermann A, Ladha JK, Olk DC, Roetter R, Tuong TP, Wade L. 1998. Research onnatural resources management: strategic research issues and IRRI’s approach to addressingthem. IRRI Discussion Paper Series No. 27. Manila (Philippines): International RiceResearch Institute. 28 p.

Pheng S, Adkins S, Olofsdotter M, Jahn GC. 1999. Allelopathic effects of rice (Oryza sativaL.) on the growth of awnless barnyard grass [Echinochloa colona (L.) Link]: a new formfor weed management. Cambodian J. Agric. 2(1):42-49.

Satake T, Yoshida S. 1978. High temperature induced sterility in indica rice at flowering. Jpn.J. Crop Sci. 47:6-17.

Savary S, Madden LV, Zadoks JC, Klein-Gebbinck HW. 1995. Use of categorical informationand correspondence analysis in plant disease epidemiology. In: Advances in botanicalresearch. Vol. 21. London: Academic Press Limited. p 213-240.

Savary S, Elazegui FA, Teng PS. 1996. A survey portfolio for the characterization of rice pestconstraints. IRRI Discussion Paper Series No. 18. Manila (Philippines): InternationalRice Research Institute. 32 p.

Sheehy JE, Mitchell PE, Beerling DJ, Tsukaguchi T, Woodward FI. 1998. Temperature of ricespikelets: thermal damage and the concept of a thermal burden. Agronomie 18:449-460.

Teng PS, Bissonnette HL. 1985. Developing equations to estimate potato yield loss caused byearly blight in Minnesota. Am. Potato J. 62:607-618.


NotesAuthors’ address: Cambodia-IRRI-Australia Project, P.O. Box 1, Phnom Penh, Cambodia.Acknowledgments: We thank Dr. Serge Savary and Dr. Paul Teng for introducing us to the data

collection and analysis techniques used in this study. We are grateful to Dr. GrahamMcClaren for advising us on the statistical analysis, to Dr. Harry Nesbitt and Dr. PeterCox for reviewing the manuscript, and to the rice farmers of Cambodia for sharing theircropping practices with us and allowing us to collect data from their fields. Financialsupport for this research was provided by AusAID.


Weed communities of gogorancah rice and reflections on management 269

Developing strategies to protect rice yields in the long term involves under-standing the structure and dynamics of weed species in response to man-agement. Where weed control is imperfect and farmers change rice cropestablishment methods and control tactics, weed shifts can occur. Further-more, in rainfed rice, the inherently complex abiotic nature of the croppingenvironment may result in variation in weed composition. A survey of theweed communities remaining after farm weeding practices was conductedduring booting of gogorancah rice (dry-seeded bunded rice) in rainfed lowlandareas of Pati and Rembang, Indonesia. Counts were made of all weed spe-cies present in four randomly placed 1-m2 quadrats at low, mid, and upperpoints of the land toposequence in fields at each of 25 farm sites. In addi-tion, soil nutrient status (pH, N, P, K, and organic matter) at each site wasmeasured.

Fifty-six weed species covering 18 families were recorded. The averagetotal weed density was 175 plants m–2, with the greatest number of speciesoccurring in upper toposequence locations. Weed communities remainingafter farmer weeding at the upper and mid positions of the toposequencewere broadly similar in species composition (Lindernia species, Echinochloacolona, Fimbristylis miliacea, and Murdannia nudiflora). These differed fromthose at the base of the toposequence, which was dominated by Ammanniabaccifera, E. colona, F. miliacea, and Leptochloa chinensis. Cyperus specieswere also abundant across the toposequence, but differed in relation toposition. Cyperus tenuispica at the top was replaced by C. iria in the middle,and, at the lower points, C. difformis was predominant. L. chinensis, a com-petitive grass weed, was also abundant in sites at the bottom of thetoposequence. Canonical correspondence analysis was used to examine in-terrelationships among sampling sites based on species composition andnutrient status. Sites at the base of the toposequence were delineated sharplyfrom those in the mid and upper positions, in which there was greater simi-larity in weed flora. Multivariate analysis showed that sample sites differed

Weed communities of gogorancah riceand reflections on managementH. Pane, E. Sutisna Noor, M. Dizon, and A.M. Mortimer

270 Pane et al

in soil nutritional status, especially for P and pH, which in turn was reflectedin species composition. The results indicate that, under current weed man-agement practices, the residual weed flora is strongly governed by hydrologi-cal factors with respect to toposequence in addition to soil factors. Futureresearch and weed management options are discussed in the context of thisbaseline characterization of weed communities.

A defining characteristic of the agroecosystem in which rainfed lowland rice is grownis the occurrence of noncontinuous flooding of land of variable depth and duration(Zeigler and Puckridge 1995). The Jakenan region in Central Java, Indonesia, with anaverage of <1,500 mm of annual rainfall, falls into the shallow drought-pronesubecosystem of the rainfed lowlands (Khush 1984). Governed by erratic seasonalrainfall, a sequence of cropping is practiced. Dry-seeded bunded rice (gogorancah) isgrown from the onset of the wet season (October-November) through to January-February, followed by minimum-tillage transplanted rice (walik jerami) through toMay, with a secondary crop (palawija) such as mungbean, maize, soybean, or cow-pea grown for the remainder of the dry season. Walik jerami and palawija crops re-main at risk from drought (Fagi 1995) and the use of on-farm water reservoirs (embung)has been actively promoted (Syamsiah et al 1994).

Rice is grown in bunded fields on sloping lands with up to a 30-degree slopeand farmers manage cropping in relation to the toposequence (Fujisaka et al 1993).As in other rainfed environments, considerable small-scale spatial variability occursin the cropping environment with respect to hydrologic processes (driven by topogra-phy, rainfall, and soil texture) and soil fertility (Wade et al 1999a). Moreover, theweed flora in these areas is potentially very diverse since the crop habitat may exhibitsoils that range from aerobic through saturated to fully flooded for varying parts ofthe crop cycle. Typically, during the first three to four weeks (depending on rainfall)of gogorancah, rice grows as an upland crop in moist soil only to be flooded as rain-fall intensity increases, as a result of impounded water. The time to flooding and itsdepth and duration in relation to field surface topography are well-known hydrologicdeterminants governing germination and seedling establishment of rice weeds andthese act interspecifically as a sieve in the recruitment of weeds into the growing crop(Pons 1982, Pane and Mansor 1994). In rainfed lowlands, these weed species areoften loosely described as “semiaquatic” in that they possess life history characteris-tics that enable establishment in moist or saturated conditions and later survival underflooded conditions, for example, many sedge species. The weed flora of walik jeramioften exhibits species in common with gogorancah but typically includes obligateaquatic species.

Weed control in gogorancah and walik jerami rice continues to rely on manualweeding. In gogorancah, weeding with a small hoe begins at the three-leaf stage ofrice about 2 wk after emergence (WAE) with further weedings at 5–6 WAE and some-


times at 7–8 WAE, 80 labor-days ha–1 being the estimated effort (Fagi 1995). In walikjerami, the crop is usually handweeded only twice, 2–4 wk after transplanting (WAT)and at 6–7 WAT, needing 48 labor-days ha–1. The extent to which weeds are a con-straint to yields has not been accurately determined by on-farm yield gap studies ingogorancah rice but a rapid rural appraisal of fields (Fujisaka et al 1993) suggestedthat weed infestations were higher in gogorancah than in walik jerami and severest inthe low-lying areas. Recently, from three seasons of research-station trials, Bangun etal (1998a,b) estimated that grain yield reductions due to weeds in gogorancah andwalik jerami were about 76% and 45%, respectively, when comparing unweeded plotswith weed-free checks.

Improving weed management practices for gogorancah rice invokes many ofthe same issues associated with direct-seeded rice production systems in otheragroecosystems (Mortimer et al 1997, Mortimer and Hill 1999). In the context ofprevailing socioeconomic production domains (Pandey 1998) and the heterogeneousenvironment of rainfed rice, the appropriate integration of agronomic and water man-agement practices to ensure rapid crop establishment of a competitive crop stand andfertilizer management and weed control interventions after crop establishment re-main important adaptive research issues. Equally essential is an understanding ofexisting farm weed management systems and the agronomic and economic constraintsexperienced by farmers. Documentation of the efficacy of farm weed control prac-tices and of the abundance and diversity of weeds throughout crop growth contributesto a baseline in the analysis of the impact of existing weed control systems and to theex ante assessment of proposed changes in management. Given a priori knowledgeabout the relative competitiveness and ecology of individual weed species, this baselineprovides a precursor to the design of on-farm yield gap trials, which are expensive toconduct (Moody 1993). Of particular importance is knowledge of the weed flora re-maining after normal farmer weeding practices.

Apart from general inventories of the weed species present in rainfed rice inIndonesia (e.g., Soerjani et al 1987), no systematic surveys have been conducted tocharacterize the on-farm weed flora of gogorancah rice. The objectives of this studywere therefore to quantitatively describe the weed flora in dry direct-seededgogorancah rice in Central Java from extensive farmer field surveys, to analyze varia-tion in this flora in relation to toposequence and soil characteristics, and to reviewimplications for improved weed management.


Survey sitesTwenty-five farm sites were chosen in subdistricts within 50 km of the Jakenan ex-perimental station of the Central Research Institute for Food Crops in the districts ofPati and Rembang, Central Java, Indonesia. Farms were identified as being in areasof intensive rainfed lowland rice production by the presence of a water reservoir inthe local vicinity and all showed sloping lands (5–30°). At each farm, three positions(upper, mid, and lower) on the toposequence were identified for census of the weed

272 Pane et al

flora. The range in elevation across the toposequence at all study sites did not exceed10 m. Prior to data collection, farmers were briefly interviewed to confirm that riceweeding had been finished and to ascertain nutrient management practices.

Weed floraDuring February 1998, crops were inspected and the weed flora assessed when ricewas at the booting stage, approximately 60–70 d after sowing (DAS). Four 1-m2

quadrats were placed randomly within the crop at each toposequence position and thedensity of individual weed species (plants m–2) enumerated by destructive removal ofall plants beyond the small seedling stage. Identification to the species level was notachieved for the genus Lindernia because of the lack of floral states. Consequently,Lindernia spp. refers in the text to several species within the genus and is likely toinclude the annuals L. anagallis, L. antipoda, and L. cillata but excludes the perennialL. crustacea, which was readily identifiable.

SoilAt each toposequence location, bulk samples were taken by pooling soil from eachindividual quadrat used for weed sampling. Two replicate sets were obtained in a likemanner. Soil parameters, including pH, were measured following standard laboratorypractices as described by Hidayat (1978). Percent total organic carbon in soil wasmeasured spectrophotometrically, % total N by semimicro Kjeldahl, soluble phos-phorus (mg kg–1) and exchangeable K (meq 100 g–1) using Bray’s method, and cationexchange capacity, CEC (meq 100 g–1), by Schollenberger’s semimicro percolationmethod.

Farmyard manure was the principal source of fertilizer although some farmersreferred to the use of inorganic phosphate and potassium.

Data analysisUnivariate analysis of variance was used to explore variation in weed densities and insoil nutrient status. Since sites were randomly chosen, and toposequence positionswithin sites were fixed, a mixed model was used, partitioning variation amongtoposequence positions within individual sites (SAS 1986).

A range of quantitative approaches was employed to explore weed communitystructure in relation to the environmental variables measured at each site. Log rankabundance curves (Ludwig and Reynolds 1989) were used to evaluate overall com-munity composition in relation to toposequence, pooling data over sites. These werethen compared with rankings given by the summed dominance ratio (Kent and Coker1992).

Ordination techniques (ter Braak 1985) were applied to quantify the weed com-munity structure in relation to soil parameters and position on the toposequence. Simpleordination methods (sensu Bray and Curtis 1957) provide graphical representationsof community structure based on either the similarity of sampled sites with respect torelative abundance of constituent species (site ordinations) or the similarity of spe-cies with respect to co-occurrence at the same sampling sites (species ordinations).


As such, they provide a method of indirect gradient analysis in that interrelationshipsbetween species and sites can be simultaneously quantified and may be interpreted inthe light of what is known, or can be inferred, about environmental gradients re-flected by relative species abundance across the range of sites. A contrasting approachis direct gradient analysis by regression methods (Whittaker 1967), which seeks toexplicitly examine a species response to chosen environmental variables or to cali-brate sites with respect to environmental variables. Indirect gradient analysis is ar-gued to have several advantages for heterogeneous environments (Jongman et al 1995)in that (1) species compositions at sites are readily identifiable and plant species areintrinsically “phytometers” of their environment, (2) environments are difficult tocharacterize exhaustively in terms of both biotic and abiotic variables, and (3) theactual occurrence of an individual species may be too unpredictable to discover itsrelationship with environmental regimes directly and general patterns of coincidenceof several species may be of greater use in detecting species-environment relations.Nevertheless, indirect gradient analysis remains intrinsically a correlative methodand can only suggest hypotheses of causal factors. Canonical ordination techniquescombine both ordination and regression into multivariate direct gradient analysis tosimultaneously interpret the response of many species to many environmental vari-ables.

We used correspondence analysis (CA) and canonical correspondence analysis(CCA) (ter Braak and Smilauer 1998) for indirect and direct gradient analysis, re-spectively, assuming species response to environmental variables on a unimodal(Gaussian) model. Data were individual species counts per m2 averaged over repli-cates at each sampling location and associated soil parameters after standardizing tocomparable scales. Log-linear regression was used to relate species abundance toenvironmental scores derived from ordination using the predictive functionexp(b0 + b1x + b2x2), with the constraint b2 <0.

Results

SoilTables 1 and 2 give soil site characteristics. Significant intersite variation was foundin mean levels of potassium, phosphate, organic carbon, and CEC, the latter beingsignificantly correlated with all variables except phosphate. No differences in nitro-gen status were found among sites or among locations within sites. In 22 out of the 25sites, statistically significant differences in soil pH were detected among toposequencepositions. Soils from the upper positions were always more acid, with the averagedifference within a site being 0.65 of a unit from upper to lower position. Contrastingly,potassium, phosphate, organic carbon, and CEC varied among locations at every farmsite and differences were not correlated with toposequence position. Levels of thesenutrients fell within reported ranges for other rainfed lowland environments, sitesbeing characteristically nutrient-deficient (Wade et al 1999b).

274 Pane et al

Tabl

e 1

. S

oil

char

acte

rist

ics

of t

he s

ampl

e si

tes.

Dat

a ar

e av

erag

ed o

ver

topo

sequ

ence

pos

itio

n an

d re

plic

ate.

Mea

nsfo

llow

ed b

y (n

) in

dica

te n

o si

gnifi

cant

var

iati

on a

mon

g po

siti

ons

wit

hin

site

s.

Site

Villa

ge/s

ubdi

stric

tpH

% N

Exch

ange

able

PO

rgan

ic C

CEC

a

num

ber

K

(meq

10

0 g

–1)

(m

g kg

–1)

(%)

(meq

10

0 g

–1)

1

Gra

wan

/Sum

ber

5.1

0.0

7 (

n)0

.06

63

0.0

7 0

.48

3.8

7 2

Prag

u/S

ulan

g 7

.6 0

.13 (n)

0.0

78

4.3

3 1

.29

10.6

6

3Ja

di/S

umbe

r 7

.8 0

.13

(n)

0.1

71

27

.22

1.5

5 2

5.6

3

4M

egul

ung/

Sum

ber

6.7

0.1

0 (

n)0

.03

31

4.9

7 0

.99

9

.32

5

Bog

orej

o/S

umbe

r 7

.7 0

.12

(n)

0.1

76

3.9

9 1

.34

24

.97

6

Ser

en/S

ulan

g 6

.1 0

.08

(n)

0.0

55

17

.91

0.5

4

7.3

1 7

Sin

dang

sari/

Lase

m 5

.8 0

.10

(n)

0.1

30

68

.32

0.7

6

8.8

4

8S

amar

an/P

amot

an 7

.0 (n)

0.1

1 (n)

0.0

73

31.8

7 1

.27

12.7

2 9

Prag

en/P

amot

an 7

.5 0

.11

(n)

0.0

46

25

.75

1.2

2 1

4.3

01

0La

mba

ngan

Wet

an/B

ulu

6.9

0.1

3 (n)

0.2

11

7.4

3 1

.05

14.6

21

1K

emba

ng/B

ulu

7.4

(n)

0.1

0 (n)

0.1

13

10.7

5 0

.95

13.9

01

2W

arug

unun

g/B

ulu

6.8

0.0

8 (

n)0

.08

31

2.8

5 0

.53

11

.04

13

Sek

arsa

ri/K

alio

ri 7

.0 0

.10

(n)

0.0

26

33

.84

1.0

6 1

0.6

714

Mat

eseh

/Kal

iori

6.3

0.1

1 (

n)0

.08

82

2.9

7 0

.95

11

.19

15

Mag

uan/

Kal

iori

7.8

0.1

3 (

n)0

.17

1 6

.72

1.5

3 2

4.1

41

6B

anyu

urip

/Gun

em 6

.3 0

.11

(n)

0.1

06

9.5

0 1

.11

15

.06

17

Pano

han/

Sul

ang

7.2

0.1

5 (

n)0

.04

8 4

.71

0.9

8 1

8.1

51

8G

unem

/Gun

em 6

.8 0

.12

(n)

0.0

73

9.9

9 1

.40

12

.80

19

Bam

ban/

Pam

otan

7.5

0.1

0 (

n)0

.05

12

9.7

8 0

.88

14

.66

20

Gem

bol/

Pam

otan

7.4

0.1

1 (

n)0

.15

82

1.2

9 1

.34

15

.15

21

Pale

mge

de/P

ucuk

wan

gi 8

.1 0

.12

(n)

0.2

28

29

.06

1.1

2 2

2.7

32

2K

aran

grej

o/Ja

ken

6.6

0.0

9 (

n)0

.07

37

5.3

1 0

.68

8

.29

23

Sid

omul

yo/J

aken

5.0

0.2

2 (

n)0

.05

51

3.3

8 0

.56

3

.76

24

Har

uman

is/J

aken

7.5

(n)

0.1

2 (n)

0.2

66

6.2

5 1

.12

26.7

42

5M

antin

gan/

Jake

n7

.4 0

.10

(n)

0.0

48

42

.52

1.0

0

6.1

8M

axim

um v

alue

rec

orde

d4

.90

.52

0.3

75

94

.80

2.2

52

9.3

8M

inim

um v

alue

rec

orde

d8

.40

.07

0.0

20

2.0

50

.20

2.6

5

a CEC

= c

atio

n ex

chan

ge c

apac

ity.


Weed abundanceFifty-six weed species (Table 3) in total from 18 familes (Amaranthaceae, Araceae,Asteraceae, Boraginaceae, Commelinaceae, Convolvulaceae, Cyperaceae,Euphorbiaceae, Lythraceae, Marsileaceae, Molluginaceae, Onagraceae, Poaceae,Pontederiaceae, Portulacaceae, Rubiaceae, Sphenocleaceae, Scrophulariaceae) wererecorded in the census. The developmental stages of species varied considerably, mostbeing vegetative and below the crop canopy height. Species commonly observed inthe flowering stage were Eclipta alba, Echinochloa crus-galli, E. colona, Fimbristylismiliacea, and Cyperus difformis and C. rotundus.

The overall mean total weed density was 175 plants m–2 (Fig. 1) and this did notdiffer significantly in relation to toposequence when averaged across sites. Positionswere ordered lower < middle < upper in terms of mean number of species. At lowpositions, the variance in densities was least and density distributions were highlyskewed in the mid and upper regions. Very high densities (above 400 plants m–2)were always dominated by a single species. Frequency distribution of counts per m2

varied among species and between toposequence positions, both Poisson and log-normal distributions being evident in the data set. For the most abundant species,distributions were typically strongly skewed but in some instances apparently dis-junct (e.g., Ammannia baccifera, lower toposequence) (Fig. 2).

Figure 3 illustrates the rank order abundance of species in each toposequence,based on pooled counts across sites. While all communities were structured geometri-cally, common taxa had a different rank in their relative abundance in relation totoposequence (Spearman’s Rs, P ≤0.03). In the upper and mid toposequence, commu-nities were dominated by a similar group of species with Lindernia spp. the mostabundant. At the low toposequence, Lindernia was replaced by Ammannia bacciferaand in low and mid positions Leptochloa chinensis was more abundant than in theupper toposequence. Both Echinochloa colona and Fimbristylis miliacea achievedhigh rankings across all toposequence positions. Ranking species by summed domi-nance ratio (averaged relative abundance and relative frequency) promoted relativelyrare species overall (e.g., Phyllanthus niruri and Digitaria ciliaris) to a much higherrank (Fig. 4), especially evident at the low toposequence.

Figure 5 shows the biplot ordination from CA of species and sites according totoposequence. In the analysis, data were logarithmically transformed and the influ-

Table 2. Correlation (product moment) matrix of soil characteristicspooled over sites and toposequence position. Data in bold are sig-nificant (P ≤ 0.05). CEC = cation exchange capacity.

pH C N P K

C 0.7252N 0.5723 0.6151P –0.2343 –0.2572 –0.2871K 0.4421 0.3680 0.2719 –0.1623CEC 0.6982 0.6778 0.5222 –0.2636 0.6190

276 Pane et al

Fig. 1. Box plots of the number of weed species andtotal weed density in relation to toposequence. Meansare shown as horizontal lines within boxes. Data areaveraged over farm sites.

Table 3. Weed species recorded and numerical code used in Figure 3.

Codes Species Codes Species

1 Ageratum conyzoides L. 28 Fimbristylis miliacea (L.) Vahl 2 Alternanthera philoxeroides 29 Hedyotis biflora (L.) Lam.

(Mart.) Griseb. 30 Hedyotis corymbosa (L.) Lam. 3 Alternanthera sessilis (L.) 31 Heliotropium indicum L.

R. Br. ex Roem. & Schult. 32 Hymenachne acutigluma (Steud.) 4 Amaranthus dubius Mart. Gilliland 5 Ammannia baccifera L. 33 Ipomoea triloba L. 6 Borreria latifolia (Aubl.) Schum. 34 Ischaemum rugosum Salisb. 7 Brachiaria mutica (Forssk.) Stapf 35 Jussiaea repens (L.) Hara 8 Commelina benghalensis L. 36 Leptochloa chinensis (L.) Nees 9 Commelina nudiflora (L.) Brenan 37 Lindernia crustacea (L.) F. Muell.10 Cynodon dactylon (L.) Pers. 38 Lindernia spp.11 Cyperus compressus L. 39 Ludwigia hyssopifolia (G. Don) Exell12 Cyperus difformis L. 40 Ludwigia octovalvis (Jacq.) Raven13 Cyperus halpan L. 41 Marsilea crenata L.14 Cyperus iria L. 42 Mollugo pentaphylla L.15 Cyperus kyllingia Endl. 43 Monochoria vaginalis (Burm. f.)16 Cyperus rotundus L. Presl17 Cyperus tenuispica Steud. 44 Murdannia nudiflora (L.) Brenan18 Dactyloctenium aegyptium (L.) . 45 Panicum maximum Jacq.

Willd 46 Paspalum distichum L.19 Digitaria ciliaris (Retz.) Koel. 47 Phyllanthus niruri Webster20 Digitaria longiflora (Retz.) Pers. 48 Phyllanthus virgatus Forst. f.21 Echinochloa colona (L.) Link 49 Polytrias amaura (Buse) O.K.22 Echinochloa crus-galli (L.) P. Beauv. 50 Portulaca oleracea L.23 Eclipta alba (L.) Hassk. 51 Scirpus juncoides Roxb.24 Eleocharis palustris (L.) R. Br. 52 Sphaeranthus indicus L.25 Eleusine indica (L.) Gaertn. 53 Sphenoclea zeylanica Gaertn.26 Eragrostis tenella (L.) P. Beauv. 54 Tridax procumbens L.

ex Roem. & Schult. 55 Typhonium trilobatum (L.) Schott27 Euphorbia hirta L. 56 Vernonia cinerea (L.) Less.

800

600

400

200

0

30

20

10

0

Plants m–2

Upper Middle Lower

Number of species


ence of rare species removed by discarding species with a proportional overall abun-dance (Fig. 3) of less than 0.1%. The ordination accounted for 34% of the variance inthe species data and 57% of the variance in the site data. Sites from the lower part ofthe toposequence were clearly separated from those in the mid and upper positions.Species were delineated on the first dominant axis (eigenvalue = 0.240, total inertia =1.91), typically those with high positive scores being obligate aquatic species (e.g.,Monochoria vaginalis and Marsilea crenata) and terrestrial species (e.g., Hedyotisbiflora and Eleusine indica) having negative ones. Consequently, the first axis maybe hypothesized to reflect the water status of sites during the cropping season. Spe-cies locations in relation to axis 2 are not readily interpretable from simple auteco-logical observations.

Figure 6 gives the response curves to the toposequence environment quantifiedby axis 1 scores for selected species together with the location of sites according toaxis score. High densities of Lindernia spp. were predicted for negative scores, which

Fig. 2. Frequency distributions of the abundance of selected weed species. Distributions forEchinochloa crus-galli and Fimbristylis miliacea are based on pooled data from all toposequencepositions. Data for Ammannia baccifera are given for each toposequence position. Note the differ-ent scales on abscissae.

1614121086420

Ammannia baccifera—high toposequence

n = 25322824201612840

Echinochloa crus-galli n = 75

Frequency

322824201612840

Fimbristylis miliacea n = 75

100 20 30 40 50 60 70 80 90 100 >100

1614121086420

Ammannia baccifera—mid toposequence

n = 25

1614121086420

Ammannia baccifera—low toposequence

n = 25

100 20 30 40 50 60 70 80 90 100 >100Number of plants per m2

278 Pane et al

Fig. 3. Log rank abundance curves of weed communities in relation to toposequence, poolingover sites: (A) upper, (B) middle, (C) lower. The species list to the right includes all thoseoccurring at greater than 1%, and species with attached data greater than 5%. Rank order byspecies code (Table 3) is given below each abscissa.

Order of species by code:38,17,28,37,5,21,44,14,43,12,40,36,47,16,51,30,52,19,46,23,55,41,39,7,24,10,26,3,2,5,18,35,4,15,8,50,56,27,34,45,20,11,22,2,13,48,31,29,33,53,54,1.

100

10

1

0.1

0.01

0.0010 10 20 30 40 50 60

Proportional abundance (log scale)

ALindernia spp. 13.6Cyperus tenuispica 9.1Fimbristylis miliacea 8.4Lindernia crustacea 7.2Ammannia baccifera 7.1Echinochloa colona 6.9Murdania nudiflora 5.7Cyperus iriaMonochoria vaginalisCyperus difformisLudwigia octovalvisLeptochloa chinensisPhyllanthus niruriCyperus rotundusScirpus juncoidesHedyotis corymbosaSphaeranthus indicusDigitaria ciliarisPaspalum distichumEclipta albaTyphonium trilobatum

%

Lindernia spp. 24.1Echinochloa colona 14.5Fimbristylis miliacea 11.8Murdania nudiflora 6.3Cyperus tenuispica 6.3Cyperus iria 6.0Leptochloa chinensis 5.8Cyperus difformisAmmannia bacciferaDigitaria ciliarisCynodon dactylonHedyotis bifloraCyperus rotundusMonochoria vaginalis

Order of species by code:38,21,28,44,17,14,36,12,5,19,10,29,16,43,26,52,41,47,15,23,39,51,34,50,46,18,1,25,8,40,3,7,45,22,49,55,13,42,31,2,30,24,11,9,56,4,27,33,48,6.

100

10

1

0.1

0.01

0.0010 10 20 30 40 50 60

B

Order of species by code:5,21,28,12,36,41,13,14,24,43,16,40,38,42,46,23,34,26,47,8,22,3,31,10,49,50,33,32,1,8,51,19,7,39,48,45,27,29,25.

100

10

1

0.1

0.01

0.0010 10 20 30 40 50 60

C Ammannia baccifera 27.5Echinochloa colona 9.5Fimbristylis miliacea 8.6Cyperus difformis 8.3Leptochloa chinensis 8.1Marsilea crenataCyperus halpanCyperus iriaEleocharis palustrisMonochoria vaginalisCyperus rotundusLudwigia octovalvisLindernia spp.Mollugo pentaphyllaPaspalum distichumEclipta albaIschaemum rugosumEragrostis tenellaPhyllanthus niruri


Fig. 4. Abundance ranked by summed dominance ratio (following Kent and Coker 1992) ofweed species present in rice at rice booting stage. Ranks are presented for the first 12 weedspecies.

Absolute density (Di ) = total number of plants for species i in all sample quadrats.Relative density (Rdi) = Di/( ∑∑∑∑∑i Di ) ××××× 100.Absolute frequency (Fi) = (The number of quadrats in which species i occurs)/4 ××××× 100.Relative frequency (Rfi ) = Fi/∑∑∑∑∑ i Fi ××××× 100.Summed dominance ratio (SDR) = (Rdi + Rfi )/2.

in turn were associated predominantly with middle and upper sites. The converse wastrue for A. baccifera, for which predicted densities increased monotonically with in-creasing score (although note that regression was not significant, P ≤0.096).Echinochloa colona and F. miliacea had similar maximum densities, the latter havinga more restricted distribution range. Regression analysis predicted more dense standsof C. difformis at sites in the lower toposequence, the distribution of Leptochloachinensis being largely similar.

CCA (Fig. 7) was performed using a restricted set of soil parameters. N content,which did not differ among sampling locations, and CEC, which was highly corre-lated with other variables, were excluded. Seventy-three percent of the species-envi-ronment relationship was explained by the first two axes of the analysis. The first axiswas strongly correlated with soil pH (r = –0.95, recorded range 4.9–8.4) and the sec-ond axis with available phosphate (r = 0.69, recorded range 2.05–94.8 mg kg–1). Spe-cies were delineated noticeably in relation to axis 1, the inference being that soil pHmay govern species distribution. Most species (left of the diagram) had low scores,

20

18

16

14

12

10

8

6

4

2

0

Upper Middle Lower

SDR (%)

Lind

erni

a sp

p.Ec

hino

chlo

a co

lona

Fim

bris

tylis

mili

acea

Cyp

erus

ten

uisp

ica

Mur

dani

a nu

diflo

raAm

man

nia

bacc

ifera

Cyp

erus

iria

Ludw

igia

oct

oval

vis

Lept

ochl

oa c

hine

nsis

Phyl

lant

hus

niru

riC

yper

us r

otun

dus

Spa

eran

thus

indi

cus

Lind

erni

a sp

p.Ec

hino

chlo

a co

lona

Fim

bris

tylis

mili

acea

Cyp

erus

iria

Mur

dani

a nu

diflo

raPh

ylla

nthu

s ni

ruri

Dig

itaria

cili

aris

Amm

anni

a ba

ccife

raC

yper

us r

otun

dus

Ludw

igia

hys

sopi

folia

Mon

ocho

ria v

agin

alis

Cyp

erus

ten

uisp

ica

Amm

anni

a ba

ccife

raEc

hino

chlo

a co

lona

Phyl

lant

hus

niru

riFi

mbr

isty

lis m

iliac

eaD

igita

ria c

iliar

is

Cyp

erus

rot

undu

s

Lept

ochl

oa c

hine

nsis

Mar

sile

a cr

enat

a

Mol

lugo

pen

taph

ylla

Cyp

erus

iria

Cyp

erus

dif

form

is

Cyp

erus

hal

pan

280 Pane et al

which were correlated with neutral or mildly alkaline soils. On the other hand, Scirpusjuncoides, Digitaria ciliata, and Lindernia spp. had high scores associated with acidicsites. Differential responses to phosphate may be postulated. Of those species com-mon to continuously flooded conditions, Leptochloa chinensis and Marselia crenatawere more abundant at sites containing high phosphate (positive correlation betweenaxis 2 and soil phosphate content) than Cyperus iria and Murdania nudiflora. Speciesin the lower left quadrant were characteristic of communities reported to occur inproductive rainfed rice habitats (Moody 1983) and were associated with high soilorganic carbon.

Discussion

Implications for yieldWade et al (1999b) reported that rice yields in weed-free researcher-managed trials ofgogorancah rice varied between 3 and 5 t ha–1 depending on water availability andthat improved nutrient management was essential to raising yields. The highest yields

Fig. 5. Biplot diagram from correspondence analysis ordination diagram of sites andspecies.Species locations are indicated by small circles and sites by large ones. = lower, = mid, = upper sites.

1.0

–1.0

Lind_sppEcli_alb

Murd_nudHedy_corLind_cruPort_oleSpha_indHedy_bif

Erag_spEuph_hir

Eleu_indDigi_cil Phyl_ninLudw_hys

Brac_mut

Typh_spCype_rot

Cyno_dacScir_jun

Cype_ten

Mars_creMono_vag

Ludw_octCype_dif

Fimb_mil Echi_cru

Comm_benAmma_bacPasp_di

Isch_rug

Lept_chi

Alte_ses

Cype_iriEchi_col

Moll_pen

Cype_hal

Axis 1

Axis

2

–1.0 1.0


Fig. 6. Log-linear regression of density of selected species per quad-rat on the site scores of the first axis of correspondence analysis(Fig. 5) assuming a Gaussian distribution function. Axis scores them-selves were derived from the site ordination and are not those shownin Figure 5, which have been rescaled for biplot presentation. Re-gressions gave significant (P ≤ 0.05) fits to abundance distributionsfor all species except Ammannia baccifera.

were found in lower toposequence positions with differences of up to 1 t ha–1 beingrecorded. Average on-farm yields in the same locality, however, have been reportedto be significantly lower and typically less than 3 t ha–1 (Syamsiah et al 1994).

While confirming the well-known variability in soil characteristics in Java, thisstudy illustrates that surprisingly dense, diverse weed communities persisted ingogorancah rice during the latter stages of crop development. Within the restrictionsof this study and by inference only, these may be strongly structured by spatial vari-ability in soil pH, by nutrient availability governed by soil pH, and by hydrologythrough toposequence position. The literature is sparse on data describing speciesdistributional ranges in relation to soil characteristics in Indonesia. It is interesting tonote, however, that Echinochloa crus-galli achieved a mid score on axis 1 (Fig. 7)and that this species is reported to prefer neutral soils (Soerjani et al 1987). Figure 3indicates that only seven species were common (>5%) in the upper and midtoposequence with five in the lower toposequence. Acidic, nutrient-poor sites in theupper toposequence were dominated by Lindernia spp., which are commonly consid-ered to be ruderal species of open sites. Habitat specificity was reflected within Cyperus,with C. tenuispica being replaced by C. difformis in lower site locations, where

50

40

30

20

10

0

Predicted density (plants m–2)

Lindernia spp.

Echinochloa colona

Fimbristylis miliacea

Ammannia baccifera

Cyperus difformis

Leptochloachinensis

–2 –1 0 1 2 Axis 1

LowerMiddle

UpperSite distribution

282 Pane et al

Ammannia baccifera was dominant. Both Echinochloa colona and Fimbristylismiliacea showed a broad adaptation to toposequence conditions (Fig. 6), although thelatter was more abundant in mid slope.

These communities are likely to have comprised plants of varying age, sincemature individuals may have escaped manual weeding operations earlier in the life ofthe crop or be younger by virtue of recruitment after weeding has ceased. Assumingthat the last weeding was approximately 40–50 days after emergence and bootingwas one month later, then 30 days may have elapsed between the completion of manualweeding and the census. This time period is of sufficient duration for populations ofEchinochloa colona, Fimbristylis miliacea, and Lindernia spp. to develop, and forspecies with strong developmental plasticity to complete the life cycle and contributeseed to future weed infestations.

Fig. 7. Biplot diagram for species and selected soil variables from canonical corre-spondence analysis. Arrows indicate the direction of change of environmentalvariables (pH, P, K, and C) and their relationship with ordination axes. Correla-tions between measured environmental variables and axis scores are inset. Spe-cies names are truncated (see Table 3).

+3.0

Ecli_alb

Mars_creLept_chi

Moll_pen Echi_col

Isch_rug

Cype_hal

Pasp_dis

Ludw_hysErag_ten

Cype_iri Murd_nud

Typh_tri Digi_cil

Scir_jun

Cype_tenLind_spp

Eleu_ind

Comm_ben

Hedy_cor

Ludw_oct

Fimb_mil

Phyl_nirCype_rot

Cyno_dac

Cype_difAlte_sesEuph_hir

Amma_bacBrac_mut

Port_oleHedy_bif

Lind_cru

Spha_ind

Axis 1

Axis

2

–2.0 +3.0

pH –0.954 –0.277C –0.616 –0.661P –0.028 0.694K –0.516 0.156

Axis 1 Axis 2

Echi_cru

Mono_vag

P

K

pH

C

–2.5


The relatively high densities of weeds and the diversity of weed species presentmay simply reflect the fact that farmers in the region perceive little economic benefitin weed removal during the later stage of crop development, particularly as furthermanual weeding may damage the crop. Moody (1993), however, has pointed out thatadditional weeding close to rice panicle initiation will often increase final yield. Clearly,without more detailed experimentation, it is only possible to hypothesize as to likelyyield reductions resulting from these observed levels of weed infestation. The follow-ing comments are in consequence purely speculative. Field observation that mostweeds were below canopy height and the anticipated limited growth periodpostweeding suggests that weed biomass per unit area may be low and that interfer-ence with photosynthetically active radiation capture by rice will be limited. Weedspecies capable of strong plasticity in growth and late development, however, such asAmmannia baccifera, may pose a competitive threat to yield during grain filling.Anecdotally, this species is reported to increase in height and biomass as the cropmatures and as fields drain. Competition for nutrients during late crop developmentmay also occur from high densities of weeds lower in the canopy. Soils in this regionare characteristically phosphate- and potassium-limited (Clough et al, this volume)and all sites surveyed in this study were nutrient-poor. While underlying heterogene-ity in soil pH may restrict the distribution of weed species, Figure 7 suggests thatthere may be differential species preferences for phosphate. This, in turn, raises thehypothesis that competition for this nutrient may occur particularly in low toposequencepositions in which weeds of irrigated rice persist. Monochoria vaginalis, Marseliacrenata, and Scirpus juncoides are all species that have been reported to competestrongly with rice for nutrients (Soerjani et al 1987) as has Eclipta alba in upland rice(Lee and Moody 1989). The extent to which Lindernia spp. may compete for nutri-ents with rice remains open to question (not least because of the lack of formal iden-tity). However, the distribution of this group of species was correlated with acidic,nutrient-poor soils in which rice yields themselves are likely to be low. The extent towhich overall productivity governs the nature and intensity of competitive interac-tions for nutrients deserves further study.

Implications for weed managementThe wide diversity of weed species present within farm localities strongly indicatesthe potential for transient temporal and longer-term shifts in relative abundance ofweed species in relation to changes in agronomy and water and weed management.Studies of soil fertility in rainfed lowlands have clearly pointed to the importance offertilizer use in increasing rice yield. Improvement in crop nutrition and deploymentof new varieties may result in suppression of many weed species simply throughenhanced crop vigor by improved agronomic practices. Many authors (e.g., Cooperet al 1999), however, have pointed to the difficulty of breeding widely adapted ricecultivars for rainfed environments tolerant of abiotic stresses throughout the life ofthe crop.

The presence of a wide spectrum of weeds encompassing species common tointensive irrigated production systems as well as rainfed and upland environments

284 Pane et al

indicates the need to deploy broad-spectrum weed control practices that are eithereffective across all toposequence positions or location-specific. At present, time-hon-ored manual weeding satisfies either option, although it may be insufficiently em-ployed in the late season. However, replacement of manual weeding by chemicalmeans will be contingent on improved water management techniques, which mayonly be feasible in the lower toposequence. Water management, in turn, will dependcritically on drainage and supply infrastructures that may be beyond the economicand practical reach of farmers. Conversely, in the upper toposequence, rapid infiltra-tion rates and lack of water supply from lower slope reservoirs may prohibit the easyuse of water in weed control.

The existence of grass weeds such as Echinochloa crus-galli, Leptochloachinensis, and Ischaemum rugosum, and of the sedge Cyperus difformis among othersedges within the agroecosytem clearly poses a significant threat to rice intensifica-tion and underlies the importance of effective early weed control. From on-farm trialsin the midtoposequence, Bangun et al (1998a,b) concluded that the use of oxadiazon(0.5 kg ha–1 a.i. applied 1 DAS) was more effective in controlling Leptochloa chinensisand Cyperus iria than manual weeding at 21 DAS and also suppressed the growth ofDactyloctenium aegyptium and Eleusine indica by 30 DAS. By 60 DAS, however,the abundance of Echinochloa colona did not differ between weed control treatmentsand additional manual weeding was required. In the same trial with IR64, no signifi-cant effects of changing crop spacing were detected in weed suppression. Bangun andothers noted, in addition, that, in the following walik jerami crop, weed biomass waslower than in gogorancah but with an increase in grass weeds including L. chinensis,E. colona, Paspalum distichum, Ischaemum rugosum, and Isachne globosa.

Fagi (1995) concluded that, while improved weed control in gogorancah ricewould necessarily be achieved by the use of broad-spectrum herbicides in both crops,significant gains could also be made by improved land preparation prior to cropping.Typically, land for walik jerami is rapidly prepared after gogorancah harvest by strawincorporation into wet soil and rice transplanted soon after to minimize risk of late-season drought. In consequence, the opportunities for the imposition of a stale seed-bed with a nonselective herbicide or repeated tillage cycles are precluded. This pointsto the potential value of short-duration rice cultivars for use in gogorancah and walikjerami to enable a window of opportunity for improved land preparation for weedcontrol either at the start of gogorancah or between rice crops. Delayed time of plant-ing of gogorancah, however, may elevate the risk of rain damage to grain quality andharvesting difficulties. The requirement for abiotic stress tolerance in such varietiesremains fundamental.

Methodological and analytical implicationsAt first sight, the application of ecological methods in the detailed analysis of weedcommunities may seem an unnecessary adjunct to developing weed management sys-tems in which the overall goal is simply the reduction of weed biomass that interfereswith crop production. However, the inherent heterogeneity of rainfed rice ecosystemsand the temporal and spatial variation that occurs in rice yield, together with the


imperfections of existing weed control technologies, call for a clear understanding ofthe role played by weeds in limiting yields and the ecological and agronomic factorsgoverning their persistence.

Summed dominance ratio (SDR) has been widely used in the past in describingthe structure of weed communities but the approach is increasingly being replaced byalternative analytical techniques. Comparisons of Figures 3 and 4 lead to differentconclusions about the relative rankings of individual species, for example, Phyllanthusniruri and Digitaria ciliaris . The SDR is often an average of two indices: one basedon the likelihood of occurrence (presence/absence in a sampling unit) and the otherbased on the absolute abundance of individual species. In consequence, the agglom-erated index is sensitive to sampling variation in two different criteria and significantdepartures from an underlying Poisson frequency distribution. Correspondence analy-ses of the type employed here are specifically designed to analyze sparse data sets(high preponderance of zeros) and to enable downweighting of the influence of rarespecies. Such multivariate analyses are also designed to detect underlying environ-mental gradients through the analysis of multiple species responses (Jongman et al1995). Their interpretation and the construction of hypotheses of explanatory pro-cesses, however, require considerable care, as evidenced by Ammannia baccifera inthis study. Indirect gradient analysis (Fig. 5) suggested that this species was associ-ated with low toposequence regimes and species of permanently flooded fields (e.g.,Cyperus difformis). Direct gradient analysis (Fig. 7), independent of toposequence,links this species with others with which it does not frequently co-occur in the field.This paradox is more apparent than real and is readily explained by the developmen-tal growth response referred to earlier and the absence of an appropriate index captur-ing the dynamics of water regimes over the crop cycle (e.g., disappearance of pondedwater, Jearakongman et al 1995) for use in analysis.

Conclusions

As in much of rainfed rice agriculture, advances in weed management in gogorancahrice will be closely linked to improvement in crop management practices throughboth the deployment of improved lines (Wade et al 1999b) and integrated nutrient andwater management (Tuong et al 1995). While adaptive research into the opportunitiesfor early postemergence chemical control practices as a replacement for, or in addi-tion to, handweeding is important, the very nature and subtlety of rainfed environ-ments will require a much better understanding of the interaction so often describedas “water-tillage-weeds.” It is here that knowledge of the effect of agronomic pro-cesses governing early recruitment of weed species is essential in relation to the in-herent variance observed across the toposequence. Especially in the instance of therainfed lowlands, farmers’ fields are fundamentally important laboratories in whichto work and ones in which on-farm yield gap trials together with the analysis of farm-ers’ perceptions and practices in weed control remain essential adaptive research toolsto protect rice yields from weeds.

286 Pane et al

ReferencesBangun P, Pane H, Jatmiko SY, Moody K. 1998a. Studies of weed dynamics and their manage-

ment in rainfed lowland rice. Rainfed Lowland Rice Research Consortium Phase II.Final Report 1994-97, Jakenan Experimental Station. Central Research Institute for FoodCrops and International Rice Research Institute. p 59-66.

Bangun P, Pane H, Jatmiko SY, Moody K. 1998b. Weed dynamics and weed managementalternatives in gogorancah and walik jerami rices. Rainfed Lowland Rice Research Con-sortium Phase II. Final Report 1994-97, Jakenan Experimental Station. Central ResearchInstitute for Food Crops and International Rice Research Institute. p 67-79.

Bray JR, Curtis JT. 1957. An ordination of the upland forest communities of southern Wiscon-sin. Ecol. Monogr. 27:325-349.

Cooper M, Fukai S, Wade LJ. 1999. How can breeding contribute to more productive andsustainable rainfed lowland rice systems? Field Crops Res. 64:3-12.

Fagi AM. 1995. Strategies for improving rainfed lowland rice production systems in CentralJava. In: Ingram KT, editor. Rainfed lowland rice, agricultural research for high-riskenvironments. Los Baños (Philippines): International Rice Research Institute. p 189-200.

Fujisaka S, Moody K, Ingram K. 1993. A descriptive study of farming practices for dry seededrainfed lowland rice in India, Indonesia and Myanmar. Agric. Ecosyst. Environ. 45:115-128.

Hidayat A. 1978. Methods of soil chemical analysis. Framework Report of the Indonesia-JapanJoint Food Crop Research Program. Japan International Cooperation Agency, Bogor,Indonesia.

Jearakongman S, Rajataserrkul S, Naklang K, Romyen P, Fukai S, Skulkhu E, Jumpake B,Nathabutr K. 1995. Growth and grain yield of contracting rice cultivars grown underdifferent conditions of water availability. Field Crops Res. 44:139-150.

Jongman RHG, ter Braak CJF, Van Tongeren OFR. 1995 Data analysis in community andlandscape ecology. Cambridge (UK): Cambridge University Press. 299 p.

Kent M, Coker P. 1992. Vegetation description and analysis. London (UK): CRC BelhavenPress. 363 p.

Khush GS. 1984. Terminology for rice growing environments In: Terminology for rice grow-ing environments. Los Baños (Philippines): International Rice Research Institute. p 5-10.

Lee HK, Moody K. 1989. Nitrogen fertilizer level on competition between upland rice andEclipta prostrata (L.) L. In: Proceedings of the 12th Asian-Pacific Weed Science Soci-ety Conference, Seoul, Korea. p 187-193.

Ludwig JA, Reynolds JF. 1989. Statistical ecology. New York (USA): Wiley. 337 p.Moody K. 1993. Weed management in rice. In: Pimentel D, editor. Handbook of pest manage-

ment in agriculture. Boca Raton, Fla. (USA): CRC Press Inc. p 301-328.Mortimer AM, Lubigan R, Piggin C. 1997. Constraints and opportunities for weed manage-

ment in rainfed lowland rice. Brighton Crop Protection Conference (1997) 2:191:196.Mortimer M, Hill JE. 1999. Weed species shifts in response to broad spectrum herbicides in

sub-tropical and tropical crops. Brighton Crop Protection Conference (1999) 2:425-437.Pandey S. 1998. Nutrient management technologies for rainfed rice in tomorrow’s Asia: eco-

nomic and institutional considerations. In: Ladha JK, Wade L, Dobermann A, ReichardtW, Kirk GJD, Piggin C, editors. Rainfed lowland rice: advances in nutrient managementresearch. Manila (Philippines): International Rice Research Institute. p 3-28.


Pane H, Mansor M. 1994. The ecology of Leptochloa chinensis (L.) Nees and its management.In: Sastroutomo SS, Auld BA, editors. Appropriate weed control in Southeast Asia. KualaLumpur (Malaysia): CABI, CAB International Regional Office for Asia. p 52-63.

Pons TL. 1982. Factors affecting weed seed germination and seedling growth in lowland ricein Indonesia. Weed Res. 22:155-161.

SAS. 1986. SAS user’s guide, statistics. Cary, N.C. (USA): SAS Institute Inc.Soerjani M, Kostermans AJGH, Tjitrosoepomo G. 1987. Weeds of rice in Indonesia. Jakarta

(Indonesia): Balai Pustaka. 716 p.Syamsiah I, Suprapto, Fagi AM, Bhuiyan SI. 1994. Collecting and conserving rainwater to

alleviate drought in rainfed ricelands of Indonesia. In: Bhuiyan SI, editor. On-farm res-ervoir systems for rainfed ricelands. Manila (Philippines): International Rice ResearchInstitute. p 141-152.

ter Braak CJF. 1985. Canonical correspondence analysis: a new eigenvector technique formultivariate direct gradient analysis. Ecology 67:1167-1179.

ter Braak CJF, Smilauer P. 1998. Canoco reference manual and user’s guide to Canoco forWindows: software for canonical community ordination. Ithaca, N.Y. (USA): Micro-computer Power. 352 p.

Tuong TP, Ingram KT, Siopongco J, Confesor RB, Boling A, Singh U, Wopereis MCS. 1995.Performance of dry seeded rainfed lowland rice in response to agrohydrology and Nfertiliser management. In: Ingram KT, editor. Rainfed lowland rice, agricultural researchfor high-risk environments. Los Baños (Philippines): International Rice Research Insti-tute. p 141-156.

Wade LJ, Fukai S, Samson BK, Ali A, Mazid MA. 1999a. Rainfed lowland rice: physicalenvironment and cultivar requirements. Field Crops Res. 64:199-210.

Wade LJ, Amarante ST, Olea A, Harnpichitvitaya D, Naklang K, Wihardjaka A, Sengar SS,Mazid MA, Singh G, McLaren CG. 1999b. Nutrient requirements in rainfed lowlandrice. Field Crops Res. 64:91-107.

Whittaker RH. 1967. Gradient analysis of vegetation. Biol. Rev. 49:207-264.Zeigler RS, Puckridge DW. 1995. Improving sustainable productivity in rice-based rainfed

lowland systems of South and Southeast Asia—feeding 4 billion people: the challengefor rice research in the 21st century. GeoJournal 35:307-324.

NotesAuthors’ addresses: H. Pane, E. Sutisna Noor, Research Institute for Rice, Sukamandi, Subang

41256, West Java; M. Dizon, A.M. Mortimer, International Rice Research Institute, LosBaños, Philippines.

Acknowledgments: We are grateful for discussions with Dr. Sunendar Kartaatmadja (CRIFC)and Dr. Mahyuddin Syam, Dr. T.P. Tuong, Dr. L. Wade, and Mr. R. Lubigan (IRRI).


Socioeconomic characterization

290 Joshi and Suresh

The role of characterization in ex ante assessment . . . 291

In a scarce and dwindling research resource scenario, characterizing produc-tion environments is a prerequisite for allocating resources more efficientlyand developing a demand-driven research agenda for wider impact. This studyintends to illustrate how characterization of a production system can facili-tate efficient allocation of research resources. The objectives are twofold:(1) characterize the rainfed rice production system and identify existing andpotential production constraints, and (2) assess research programs in an exante framework to allocate resources more efficiently and meet nationalobjectives. The study has been undertaken in the rainfed rice productionsystem, which is largely confined to eastern India. This production systemlagged far behind the Green Revolution belt in agricultural development. Ithas now been recognized that future sources of agricultural growth lie in therainfed rice production system. Therefore, investment in agricultural resourcesshould be able to tap the potential of this production system. It has beencharacterized based on its agroclimatic features and the economic contribu-tion of important enterprises. Characterization has aimed to delineate a ho-mogeneous production environment to better understand common produc-tion constraints, identify technological options to alleviate these constraintsin a target domain, and accelerate adoption to increase the impact of re-search resources. Five criteria were used to assess potential technologicaloptions in the ex ante framework: efficiency, household food security, genderissues, crop diversification, and sustainability of natural resources. Thesecriteria have been considered important for their contribution to meeting thesocioeconomic and environmental objectives in the rainfed system. The im-pact of efficiency has been measured using the economic surplus approachand quantifying the net present value and internal rate of returns for eachtechnology option. Other criteria have been assigned ranks between 1 and 5depending on their contribution. The ex ante assessment of various pro-grams and technological options in the rainfed production system noted thatthe research agenda was biased in favor of a few commodities and ignoredsome important and potential activities. Reallocation of research resourceshas been proposed to develop demand-driven technological choices. Thecharacterization of the production system also helped to identify niches fordisseminating potential technological options in the target domain.

The role of characterizationin ex ante assessment of researchprograms: a study in the rainfed riceproduction systemP.K. Joshi and Suresh Pal

292 Joshi and Suresh Pal

Characterization of production environments in agriculture is gaining considerableimportance for identifying the major constraints to production and technology adop-tion (ICRISAT 1998, CRIDA 1998). Characterization of production environments isusually undertaken to understand agroclimatic environments, resource endowments,and production and socioeconomic constraints to be able to identify and prioritizeresearch programs. Delineating homogeneous production environments makes it pos-sible to assess research capacity and gaps. This helps in designing need-based tech-nologies, which are expected to reduce research and adoption lags. This enhancesagricultural research efficiency and accelerates the impact of investments in research.

In the past, several characterization studies attempted to understand the targetresearch domains, mainly with respect to climate, insect pests, soils, etc. Althoughcharacterization information was largely used to delineate homogeneous agro-ecoregions and production environments, and to a lesser extent design technologiesfor alleviating one or more production constraints, it was not further applied to assessthe feasibility of research programs. With shrinking research resources and increas-ingly complex problems in agriculture, there is a need to use characterization infor-mation more rigorously to assess research programs and improve research efficiency.In the past, increasing food-grain production to meet self-sufficiency was the majortarget of agricultural research.

The new set of problems has broadened the agricultural research focus to ad-dress the sustainability of natural resources, the conservation of biodiversity, and otherareas. Publicly funded research has to meet multiple goals (e.g., equity, sustainability,food and nutritional security, diversification) in addition to efficiency issues for so-cial welfare. The new paradigm therefore calls for effectively using characterizationinformation to better target and plan agricultural research programs.

This chapter attempts to use characterization information for an ex ante assess-ment of research programs to assist in the judicious use of scarce research resources,keeping in view the social objectives. The study focuses on the rainfed rice produc-tion system in India, a system characterized by low productivity, slow and poor dis-semination of new technologies, a large concentration of poor people, high degrada-tion of natural resources, including biodiversity, and poor infrastructure. Althoughthis system is lagging far behind the irrigated and other favorable regions, consider-able potential and opportunities exist, as it possesses fairly good soils, high precipita-tion, enough human resources, and a large cattle population (Joshi et al 1999).

The study is based on the projects submitted to the rainfed rice production sys-tem in the rainfed ecoregion under the National Agricultural Technology Project(NATP). The World Bank approved a sum of US$239.70 million in 1998 to strengthenagricultural research in India. The World Bank and the government of India are shar-ing the cost of the project. Its principal objective is to address the key constraints thatcurrently limit, and which if not addressed would in the future increasingly restrict,the efficient use of the public resources that India devotes to the generation, assess-ment, and transfer of agricultural technology. To facilitate the process, the country isdivided into five ecoregions: arid, coastal, hills and mountains, irrigated, and rainfed.The present exercise was done for the rainfed ecoregion.


Approach

The following steps were taken to assess research programs in an ex ante frameworkusing characterization information:

Step I: Delineation of the rainfed rice production systemTo undertake more focused research in the rainfed rice production system, it wasnecessary to identify the research domain, which represents the predominant rainfedrice area. For this purpose, the data (1990-94 series) generated in a project on “Sus-tainable Rainfed Agricultural Research and Development” were used (ICRISAT 1998).The rainfed rice production system in India is delineated into different ecoregions asfollows:

1. Agroecological subregions from 3 to 13 (delineated by the National Bureauof Soil Survey and Land Use Planning) were included because the remain-ing subregions fall under different ecoregions, such as arid, hills and moun-tains, irrigated, and coastal. This step identified 280 districts.

2. Districts having less than 40% irrigated area were selected in the secondstage. This step reduced the number of districts to 152.

3. Districts having rice area more than 20% of the gross cropped area wereretained to focus on rainfed rice. This yielded a list of 50 districts.

4. To maintain contiguity of districts, three (two in Uttar Pradesh and one inMaharashtra) were eliminated. This step confined the cluster to 47 districts,characterized as the rainfed rice production system.

The districts identified in step 5 cover about 85% (about 10 million ha) of thetotal rainfed rice area in the country. The average yield of these districts is nearly 1 tha–1.

Step II: Identification of constraints and technologiesIdeally, the following steps should be adopted:

1. Identify and prioritize production constraints based on yield loss, extent ofthe constraints, and probability of their occurrence.

2. Identify possible technologies to alleviate constraints.3. Develop research programs in case location-specific technologies are not

available.This step is largely based on the information generated during characterization.

Based on available information and scientists’ perceptions, constraints were identi-fied in the rainfed rice production system.

Step III: Compilation of minimum data setThe costs of research were estimated. To assess the potential benefits as a result oftechnology intervention, the following information was compiled:

A. Technical informationi. Existing yields of crops under studyii. Expected yields as a result of technology intervention


iii. Unit cost reductioniv. Change in resource-use patternv. Characteristics of improved technology (i.e., technology traits)vi. Base level of area and production of major crops in the target domainvii. Fallow land

B. Socioeconomic informationi. Input and output pricesii. Demand and supply elasticities. These are important for assessing how

farmers and consumers respond to changing prices.iii. Population below poverty lineiv. Percentage of literate farm women

C. Research processi. Research lag. This is the time difference between the starting year of

the research project and when the research output (i.e., technology) isidentified.

ii. Probability of success. This is the probability of success expected inachieving the objectives set in the target period.

iii. Adoption of technology. This is the expected adoption and ceiling levelof adoption in the target domain.

Data sets A and B were based on the information generated to characterize theproduction system, and the research team. Data set C was collected from the researchteam involved in developing technology. The information supplied by the researchteams was discussed with specialists and extension agents, and some modificationswere made based on their past experiences. More discussion focused on the probabil-ity of success, which largely depends on the strength of the research station in termsof facilities and human resources.

Step IV: Assessment of benefitsThe economic surplus approach was used to assess the potential benefits generated asa result of the technology intervention. The approach is used with the assumption thattechnology intervention would improve supply, reduce the unit cost of production,and benefit consumers and producers. Figure 1 gives a simple, conventional, com-parative-static partial equilibrium model of supply and demand in a commodity mar-ket. DD is the demand curve for the commodity under study. S0S0 is the supply curveof the commodity under study before the technology intervention. With DD demandfor a commodity and S0S0 supply of the commodity, the equilibrium price would beP0 and the quantity Q0. With technology intervention, the production function wouldshift upward and the unit cost of production would come down. Under this scenario,the supply curve of the commodity would shift to the right-hand side. The new supplycurve would be S1S1. With the new supply function, the equilibrium price would beP1 and the quantity Q1. Prices would fall and quantity supply and demand would behigher at the new equilibrium. If prices fell, consumers would always be the gainers.But producers would be losers as a result of the fall in prices, but gainers due to the


increase in supply. The net gain to producers would depend on whether the increasein production compensates for the fall in prices.

Yield-enhancement or cost-reduction technology intervention would result in achange in the consumer’s surplus equal to the area P0abP1. Similarly, the producer’ssurplus would be represented by the area P1bS1 – P0aS0 (approximately P1bcd). Thetotal economic surplus would be equal to the area S0abS1. This economic surplus isadjusted with the expected adoption and probability of success. Economic surplus isestimated for each research program in the rainfed rice production system. This infor-mation is used to assess the net present value (NPV) of each research program asfollows:

n

NPV = ∑ [(ES * Ps * ADi – RCi]/(1 + r)i

i = 1

where ES is the economic surplus, Ps is the probability of success, ADi is the adoptionof the technology in the ith year, RCi is the research cost in the ith year, r is thediscount rate, and i is the time period.

The internal rate of return (IRR) of each project was also computed. To com-pute NPV and IRR, the following assumptions were made:

● Supply and demand elasticities of different commodities were used fromKumar (1997). These are important for understanding the response of farm-ers and consumers in the event of changing prices and supply.

● Adoption of improved technologies as a result of research under the NATPwas considered up to 2020 with technology degeneration at a linear rateafter reaching the ceiling level.

Fig. 1. A framework for measuring producer andconsumer surpluses.

Price

D

0 Q0 Quantity

S0

S1

a

b

cS0

S1

P1

P0

D

Q1

d


● 1997-98 was used as the base year (i.e., start of the research project) for thetarget domain of the improved technologies and output prices.

● The target domain of the improved technologies was assumed to be the eco-logical region of the research station/center(s).

● The economic surplus approach in a closed economy model was used toestimate the total economic surplus and NPV.

Step V: Assessment of other indicatorsIn addition to the net present value (which represents the efficiency indicator), fiveother criteria were used: sustainability, household food security, gender, trade, andcrop diversification. Since quantitative information for these criteria was not avail-able, these were subjectively assigned on a scale of 1–5 (with 1 the lowest and 5 themaximum) based on the contribution of the research program.

Step VI: Prioritizing research programsA composite index of all the criteria (efficiency, sustainability, household food secu-rity, gender, trade, and crop diversification) was developed by assigning appropriateweights to the selected indicators. The efficiency indicator was assigned a weight of0.50, food security 0.20, and gender issues, sustainability, and crop diversification0.10 each. Weights were decided in consultation with the scientific community andwere based on the significance of the criteria in the target domain. The purpose ofdeveloping composite indices was to rank all the projects of the proposed rainfed riceresearch program in view of their expected contribution toward multiple objectives.

Step VII: Prioritization across commodities at the production systems levelThe above steps were followed to prioritize research programs to be implemented inthe target domain. This is a bottom-up approach to prioritize research programs.Prioritization across commodities at the production systems level was also done tomatch the national and regional priorities. For this, the congruence approach wasused, along with three criteria: (1) efficiency (measured as the value of crop output),(2) equity (measured as illiterate farm women in the target domain), and (3)sustainability (fallow land was taken as the proxy for degradation of natural resources).

Prioritization of research programs

Twenty research projects in the rainfed rice program under the NATP were assessedin an ex ante framework. NPV and IRR were computed using the economic surplusapproach. To compute NPV and IRR, data pertaining to yield and cost of cultivationof the existing best technology and of proposed research were collected from theresearch teams.

Table 1 contains the results of the efficiency indicator based on the ex anteassessment. The expected IRR and NPV of all the research programs were very high,which suggested a high potential for research on generating economic surplus. Thetop five projects generating the highest economic surplus as a result of research suc-


cess addressed low yields, drought management, weed management, diseases in live-stock, and nutrient deficiencies in the livestock system. The next five projects fo-cused on integrated nutrient management (INM), soil degradation, integrated pestmanagement (IPM), management of excess water, and INM in vertisols and alfisols.

In the next stage, the efficiency indicator was complemented by food security,equity/gender issues, sustainability, and crop diversification. The projects were as-sessed on the basis of their expected contribution to improving food security, meetingthe needs of women in agriculture, enhancing the sustainability of natural resources,

Table 1. Ex ante assessment of research programs of the rainfed rice productionsystem.

NPVa IRRb

Research project Constraint (million (%) RankUS$)

Improve crop yield ceiling Low yield 639.30 114 1Rainwater management for Drought 277.55 184 2

drought alleviationWeed management Weeds 241.65 190 3Control of parasitic diseases Diseases 182.80 197 4Sustainable livestock Nutrient deficiency 117.85 98 5

production systemsIntegrated plant nutrient Low organic matter, 114.48 142 6

management N and P deficiencySoil quality and degradation Erosion, nutrient 85.10 252 7

deficiencyIntegrated pest management Pest damage 77.88 79 8Managing excess water Excess water 63.33 125 9INMc in vertisols and alfisols Low N and P efficiency 54.08 197 10Soil tillage guidelines Crusting 53.65 109 11INM in fish cultivation Manure unavailability 41.30 167 12Nutrient management of Low yields 39.98 141 13

hybrid riceVegetable-based production Fallow lands 36.50 237 14

systemsImpact of tank irrigation Lack of assured water 28.33 183 15Restoration of degraded Degradation, runoff 20.35 90 16

watershedsBioinoculants Low rhizobium 17.63 103 17Characterization of rainfed Inappropriate diagnosis 16.95 112 18

rice systemDynamics of socioeconomic Inappropriate diagnosis 8.63 77 19

changesCrop management strategies Fallow lands 2.45 46 20

to increase croppingintensity

aNPV = net present value converted into US$ using the exchange rate of US$1 = Rs 40. bIRR =internal rate of return. cINM = integrated nutrient management.


Tabl

e 2. A

sses

smen

t of

res

earc

h pr

ogra

ms

base

d on

mul

tipl

e go

als.

Ran

ked

byR

esea

rch

Cum

ulat

ive

NPV

aIR

RFo

odS

usta

in-

Gen

der

Cro

pco

stco

stPr

ojec

tC

onst

rain

t(m

illio

n(%

)se

curit

ybab

ility

bis

sues

bdi

vers

ifi-

Effic

ienc

yC

ompo

site

US

$)

catio

nb in

dex

inde

x(m

illio

n U

S$

)

Impr

ove

crop

yie

ld c

eilin

gLo

w y

ield

63

9.3

01

14

43

31

11

0.1

91

50

.19

15

Rai

nwat

er m

anag

emen

t fo

rD

roug

ht2

77

.55

18

45

43

22

20

.36

05

0.5

52

0dr

ough

t al

levi

atio

nC

ontr

ol o

f pa

rasi

tic d

isea

ses

Dis

ease

s1

82

.80

19

74

33

44

30

.36

03

0.9

12

3S

usta

inab

le li

vest

ock

prod

uctio

nN

utrie

nt1

17

.85

98

44

34

94

0.4

07

01

.31

93

syst

ems

defic

ienc

yM

anag

ing

exce

ss w

ater

Exce

ss w

ater

63

.33

12

54

53

48

50

.19

85

1.5

17

8W

eed

man

agem

ent

Wee

ds2

41

.65

19

03

33

13

60

.39

25

1.9

10

3S

oil q

ualit

y an

d de

grad

atio

nEr

osio

n,8

5.1

02

52

45

32

67

0.9

27

52

.00

30

nutr

ient

defic

ienc

yR

esto

ratio

n of

deg

rade

d w

ater

shed

sD

egra

datio

n,2

0.3

59

05

54

11

58

0.9

90

02

.10

20

runo

ffIn

tegr

ated

pes

t m

anag

emen

tPe

st d

amag

e7

7.8

87

94

54

17

90

.90

28

3.0

04

8C

rop

man

agem

ent

stra

tegi

es t

oFa

llow

land

s2

.45

46

45

34

18

10

0.4

68

03

.47

28

incr

ease

cro

ppin

g in

tens

ityN

utrie

nt m

anag

emen

t of

hyb

rid r

ice

Low

yie

lds

39

.98

14

15

22

11

31

10

.06

40

3.5

36

5Ve

geta

ble-

base

d pr

oduc

tion

syst

ems

Fallo

w la

nds

36

.50

23

73

33

41

41

20

.23

70

3.7

73

5In

tegr

ated

nut

rient

man

agem

ent

inM

anur

e4

1.3

01

67

32

24

12

13

0.1

01

33

.87

48

fish

cultu

reun

avai

labi

lity

Vege

tabl

e cu

ltiva

tion

and

stor

age

Fallo

w la

nds

2.2

04

83

33

42

01

40

.71

00

3.9

45

5IN

M in

ver

tisol

s an

d al

fisol

sLo

w N

and

P5

4.0

81

97

34

21

10

15

0.6

87

54

.01

43

effic

ienc

y

cont

inue

d on

nex

t pa

ge


Tabl

e 2

con

tinu

ed.

Ran

ked

byR

esea

rch

Cum

ulat

ive

NPV

aIR

RFo

odS

usta

in-

Gen

der

Cro

pco

stco

stPr

ojec

tC

onst

rain

t(m

illio

n(%

)se

curit

ybab

ility

bis

sues

bdi

vers

ifi-

Effic

ienc

yC

ompo

site

US

$)

catio

nb in

dex

inde

x(m

illio

n U

S$

)

Inte

grat

ed p

lant

nut

rient

Low

OM

, N

11

4.4

81

42

24

21

51

60

.17

68

4.1

91

0m

anag

emen

tan

d P

defic

ienc

yB

ioin

ocul

ants

Low

rhi

zobi

um1

7.6

31

03

24

22

16

17

0.2

87

34

.47

80

Appr

opria

te in

ocul

ants

Low

rhi

zobi

um4

.63

59

24

22

17

18

0.2

83

34

.76

13

Soi

l till

age

guid

elin

esC

rust

ing

53

.65

10

92

32

11

11

90

.10

70

4.8

68

5Im

pact

of

tank

irrig

atio

nLa

ck o

f as

sure

d2

8.3

31

83

12

12

19

20

0.2

00

05

.06

85

wat

er

a NPV

= n

et p

rese

nt v

alue

, IR

R =

inte

rnal

rate

of r

etur

n, O

M =

org

anic

mat

ter.

Exch

ange

rate

use

d: U

S$

1 =

Rs

40

. bO

n a

scal

e of

1–5

, whe

re 1

= th

e lo

wes

t and

5 =

the

high

est.


and improving crop diversification. Table 2 gives the results of this analysis. Theranking of projects changed when all indicators of national and regional prioritieswere considered. Increasing yield levels and improving drought management retainedthe same priorities, but the two projects dealing with the livestock system (diseasesand nutrient management) were ranked higher because of their contribution towardsustainability and diversification. Similarly, a project such as the restoration of de-graded watersheds, which was ranked 16th by efficiency criteria, moved to 8th be-cause of the contribution of research outputs toward sustainability. This is the projectthat proposed to develop technologies to use fallow lands through legumes in therice-fallow system. Although legumes contribute too little in terms of profits, theirrole in improving soil fertility and conserving soil and water resources is well recog-nized. The project was elevated because of the contribution of legumes to improvingsustainability, encouraging crop diversification in fallow lands, and achieving foodsecurity.

The ranking of research projects and cumulative research costs provided usefulinformation for decision-making. The total budget requirement of all these projectsunder the NATP was estimated to be US$5,224,000 (or Rs 208.96 million at an ex-change rate of Rs 40 = US$1). This suggested that only the top 11 projects should besupported in case $3,750,000 was available for the rainfed rice production system.The information also supports the justification for more research funds. For example,in case the last four projects are not allocated resources (which are about $877,500),the potential benefit sacrificed would be about $104,250,000. This is valuable infor-mation for research management, which suggests that raising the research budgetwould generate a huge economic surplus.

Role of characterization

Characterization played a vital role in reallocating research resources to better targetresearch to areas expected to contribute relatively more in various dimensions. In thepresent production system, if the entire research program is supported by the NATP,the budget distribution is as follows: 79% for crop production activities, includingnatural resource management, diagnostic surveys, and socioeconomic studies; 16%for animal husbandry; and 5% for horticultural research. Based on the contributionsof different enterprises, the extent of poverty in the target domain, and the degrada-tion of natural resources, there is a need to reallocate resources to maximize privateand social goals.

The analysis suggested that, within the rainfed rice production system, cropactivities should receive about 61% of resources, followed by 24% for fruits andvegetables, 12% for dairy enterprises, and 3% for small ruminants. The compositionof the research resource allocation changes in two subproduction systems, dependingon the importance of different activities (Table 3). It will not be desirable to allocateall available research resources (61%) in the rice production system to the rice crop.This is relevant because some other crops are also of economic importance to thefarming community, and these should also receive some resources, depending on


Table 3. Research resource allocation (%) within the rainfed riceecosystem.

Rainfed rice Rainfed rice AggregateActivity with fruits and with animal allocation in

vegetables husbandry rainfed ecosystem

Crop production 58 62 61Fruits and vegetables 35 16 24Dairy 5 18 12Small ruminants 2 4 3

Table 4. Research resource allocation to different crops in the rainfedrice production system.

Rainfed rice Rainfed rice Rainfed riceCrop with fruits and with animal production

vegetables husbandry system

Rice 52 49 50Maize 2 1 1Wheat 2 3 2Pigeonpea 1 0 1Rape and mustard 1 0 2Sesamum 0 3 2Groundnut 0 6 3

their significance within the production system. An exercise on research resourceallocation across crops suggested that half of the total available resources for researchon the rainfed ecosystem should go to rice research, and about 11% to other crops(Table 4).

Rice research should receive the bulk of the resources in the rainfed rice pro-duction system. The crop is grown in diverse environments. According to the eco-logical distribution of rice, research resources for lowland rice should be about 30%(Table 5). Upland rice research should receive 15% of the total research resourcesavailable in the rainfed rice ecosystem and 5% of the total should be earmarked fordeepwater rice.

The information generated during the characterization of the production sys-tem provided a valuable input for setting research priorities based on the existingproduction constraints, and for linking this information with the aggregate-level pri-ority setting, which includes national objectives as well.

Conclusions

This study focused on an ex ante evaluation of research programs in the rainfed riceproduction system under the NATP. The information generated to characterize the


Table 5. Research resource allocation (%) to riceresearch in different ecologies.

ResearchRice ecology Subecologya resource

allocation

Lowland rice Shallow-water rice 20Intermediate-water 10

riceUpland rice Upland rice 15Deepwater Semideepwater 3

rice riceDeepwater rice 2

aShallow-water rice = 0–30 cm, intermediate water =30–50 cm, semideepwater = 50–100 cm, and deepwaterrice = >100 cm.

production system was used to assess research programs for better targeting of re-search resources to meet the multiple goals of society: efficiency, food security, meet-ing the needs of women in agriculture, sustainability of natural resources, and cropdiversification. The analysis revealed that there is a need to reallocate research re-sources to maximize research efficiency and meet social and environmental objec-tives.

The information generated for characterization also suggested gaps in the exist-ing research programs. Linking the information from characterization and the ex anteassessment of research programs provided several opportunities that ought to be ex-plored through better-targeted research.

ReferencesCRIDA (Central Research Institute for Dryland Agriculture). 1998. District-based research

prioritization in the rainfed rice-based production system. Hyderabad (India): CRIDA.ICRISAT (International Crops Research Institute for the Semi-Arid Tropics). 1998. Sustain-

able rainfed agricultural research and development project: database development, ty-pology construction, and economic policy analysis. Module I. Patancheru (India):ICRISAT.

Joshi PK, Suresh Pal, Vittal KPR. 1999. Research prioritization of the rainfed rice productionsystem. In: Suresh Pal, Joshi PK, editors. New paradigms of agricultural research man-agement. New Delhi (India): National Centre for Agricultural Economics and PolicyResearch.

Kumar P. 1997. Supply demand projections. New Delhi (India): Indian Agricultural ResearchInstitute.


NotesAuthors’ address: National Centre for Agricultural Economics and Policy Research, New Delhi

110 012, India.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-


Constraints to the adoption of modern varieties of rice . . . 305

The introduction of high-yielding modern varieties (HYVs) and seed-fertilizertechnology in agriculture during the mid-1960s has led to a marked increasein the growth of agricultural output and has been instrumental in transform-ing traditional household agriculture into modern, commercial agriculture insome agriculturally developed states. The rainfed rice production system,which is largely confined to the eastern part of India, lagged far behind theGreen Revolution belt in agricultural development and prosperity. Of late, ithas been recognized that future sources of agricultural growth lie in thisregion.

The adoption of modern varieties and associated technologies seemsto offer an opportunity to increase output and income substantially. But evennow, the pattern and pace of adoption of modern rice varieties and othercomponent technologies have had only partial success in the rainfed ricesystem. Several biotic and abiotic stresses and socioeconomic and institu-tional constraints limit their adoption in the rainfed rice system. This studyaims to analyze constraints to the adoption of modern varieties and othercomponent technologies in the rainfed lowland ecosystem of Bihar. Bothcross-sectional primary data and time-series secondary data were used.

The bulk of rice is harvested in the kharif season, which suffers fromthe vagaries of the monsoon characterized by drought or floods and vulner-ability to pests and diseases. The technical constraints to the adoption ofmodern varieties and component technologies are highly variable accordingto soil-water relationships, which differ from one ecosystem to another. Thelack of tolerance for submergence, insect pests such as gall midge and brownplanthopper, and diseases such as tungro, sheath blight, and bacterial blightare likely to be important obstacles inhibiting adoption in the rainfed lowlandrice ecosystem.

Several studies have reported the unavailability of modern varieties asthe major limitation to their adoption in the rainfed rice production system.Bihar also suffers from outdated tenurial relations that adversely affect adop-tion. A related problem in the entire eastern region is the fragmentation ofholdings. There is a need to undertake tenurial reforms in a large way andsimultaneously consolidate holdings.

Constraints to the adoption of modernvarieties of rice in Bihar, eastern IndiaA. Kumar and A.K. Jha

306 Kumar and Jha

In addition to the unavailability of modern varieties, their poor threshabilityand thatchability, the requirement for additional capital and labor, excessiveweed infestation, the lack of timeliness or availability of fertilizer, and lack ofcredit along with knowledge and cultural practices may be instrumental totheir slow adoption in this region. The adoption of component technologiessuch as fertilizer and pesticides seems to have been affected by the adul-teration prevalent in the market. The input delivery system is very weak inBihar. The region also lacks research and development and appropriate ex-tension services. Poor roads, communications, and market infrastructureare also important constraints.

Addressing some of these constraints through appropriate research andpolicy intervention could have a large impact on the adoption of modernvarieties and their component technologies.

The introduction of high-yielding modern varieties popularly known as seed-fertil-izer technology in Indian agriculture during the mid-1960s has led to a marked in-crease in the growth rate of agricultural output and has been instrumental in trans-forming traditional household agriculture into modern, commercial agriculture in someof the agriculturally developed states. The rainfed rice production system, which islargely confined to the eastern part of India, lagged far behind the Green Revolutionbelt in terms of agricultural development and prosperity. Of late, it has been recog-nized that future sources of agricultural growth lie in this region.

The adoption of modern varieties (MVs) and associated technologies seems tooffer an opportunity to increase output and income substantially. But even now, thepattern and pace of adoption of MVs of rice and other component technologies havemet with only partial success, particularly in the rainfed rice system. For instance, therice area under high-yielding varieties (HYV) in Bihar is 60%, whereas it is 94% inPunjab. The conventional wisdom is that constraints to the rapid adoption of innova-tions involve factors such as the lack of credit, limited access to information, aversionto risk, inadequate farm size, inadequate incentives associated with farm tenure ar-rangements, insufficient human capital, the absence of equipment to relieve laborshortages (thus preventing timeliness of operations), the chaotic supply of comple-mentary inputs (such as seed, chemicals, and water), and inappropriate transportationinfrastructure (Feder et al 1985).

The purpose of this chapter is to provide an analysis of constraints to the adop-tion of MVs of rice in eastern India on the basis of a survey of various studies thathave attempted to explain these constraints and component technologies for MVs aswell as a study of farm-level primary data obtained from eight villages, two eachbelonging to Pusa and Kalyanpur blocks of Samastipur District in the north Biharplains and Pali and Bihta blocks of Patna District in the south Bihar plains, to makethe study more comprehensive and meaningful. Among the major rice-producing statesin eastern India, Bihar was selected because, even after more than three decades of


the Green Revolution, the coverage of HYV rice area is merely 60%, which is aheadof only Assam (50%).

This chapter has two main parts. The first section briefly reviews some of therelevant past work. Because the volume of such published research is overwhelming,we have attempted simply to review some representative work rather than to presentan exhaustive discussion of all work. The second section attempts to identify andanalyze constraints to the adoption of MVs and other component technologies in therainfed lowland ecosystem of Bihar.

To summarize the vast amount of empirical literature on adoption constraintssystematically, we organized the review of this work according to the key constraintsto adoption.

Biophysical constraints

The bulk of rice in the rainfed ecosystem is grown in the rainy (kharif) season, inwhich the vagaries of the monsoon, characterized by drought or floods, as well as thereoccurrence of other biotic and abiotic constraints are remarkable. The hot and hu-mid climate supports the outbreak of several harmful pests and diseases. Since it isnot possible to manipulate nature (climate), we need to incorporate sufficient toler-ance and adaptability into the rice varieties to be developed for the rainfed ecosystem.The lack of tolerance in rice varieties for abiotic constraints such as flash floods andfrequent drought as well as biotic constraints such as gall midge, brown planthopper,tungro, sheath blight, and bacterial blight seems to affect the adoption of modern ricevarieties and their component technologies in the rainfed rice ecosystem. Tripathi(1977) identified susceptibility of HYVs to diseases and pests, low germination, inef-fectiveness of dry seed treatment, nonavailability of irrigation water in summer forthe nursery, and serious waterlogging in the wet nursery as important constraints incoastal Orissa, whereas Gowda and Jolihal (1969) have reported the unsuitability ofMVs for late planting as an important constraint. The lack of irrigation is also one ofthe most limiting factors affecting the adoption of MVs and their associated technolo-gies.

Roy (1976) found inappropriate irrigation facilities as the most important con-straint to the adoption of HYVs in the kharif season besides disease incidence and thelack of suitable varieties in West Bengal. Shakya and Flinn (1985) examined factorsinfluencing the adoption of MVs and fertilizer in the Tarai of southeastern Nepal andstated that adoption of MVs is highest where irrigation exists. Because rice lands ineastern India will remain rainfed in the foreseeable future, a greater spread of MVsinto these adverse environments will probably depend on new varieties being bredthat are specifically adapted to these environments.

308 Kumar and Jha

Socioeconomic constraints

Farm sizeFarm size is one of the important factors on which the empirical adoption literaturefocuses (Feder et al 1985). Several studies have tried to highlight the associationbetween farm size and the adoption of HYV technology (Bhati 1976, Palmer 1976,Asaduzzaman 1979, Ahmad 1981, Misra et al 1986, Sarap and Vashist 1994) andhave found a positive relationship. Contradictory evidence, however, is also not un-common. Hayami (1981), from Barker and Herdt’s (1978) study of 30 villages in fiveAsian countries, mentioned that the relationship between the adoption of modern ricevarieties and absolute farm size for a cross-country pooled sample is negative. Muthia(1971), Schluter (1971), and Sharma (1973) found that small and medium-sized farmsin India adopted HYVs on a larger proportion of area than did large farms. Fuglie(1992) found a similar result in the rainfed rice-farming system in northeast Thailand.

In general, larger farms, because of their income, economic power, social pres-tige, and links with local political leadership, have a more assured supply of moderninputs including credit facilities necessary for fruitfully using the potential of newtechnology (Rehman 1983, Sarap and Vashist 1994). Moreover, farm size is a prereq-uisite for several factors such as access to credit, inputs, and information. As such,large farmers enjoy preferential treatment in obtaining such inputs (Mitra 1971, Thorner1964, Sarap 1990, Sarap and Vashist 1994).

Risk and uncertaintyInnovations in most cases face risk and uncertainty, which may affect adoption. Hazell(1982), however, found that variance in cereal production increased over time be-cause of other factors and not because of the adoption of HYVs. On the other hand,Walker and Ryan (1990) reported the adoption of HYVs of sorghum and pearl milletto be a major contributing factor to the increased production variability of these crops.Singh and Byerlee (1990) found variability in wheat yield, measured by the coeffi-cient of variation, to have decreased over time mainly as a result of an expansion inirrigated area. Pandey et al (2000) found that an increase in the percentage of HYVarea has a stabilizing effect on yield variability, which is contrary to the conventionalwisdom that HYVs lead to greater instability. Although one would expect irrigationto have a stabilizing effect, this may not hold true in the case of eastern India becauserice is only partially irrigated in this region and the irrigation supply is not reliable.

Credit constraintsIn spite of the view that a lack of credit alone does not restrict the adoption of tech-nologies that are scale-neutral (Schutjer and Van der Veen 1977, von Pischke 1978),differential access to capital is often cited as one of the factors in differential rates ofadoption (Feder et al 1985). Several studies found that a lack of credit significantlylimits the adoption of HYV technology even where fixed pecuniary costs are notlarge (Bhalla 1979, Frankel 1971, Wills 1972, Khan 1975, Behra and Sahoo 1975,Shakya and Flinn 1985, Sarap 1990, Fuglie 1992, Green and Ng’ong’ola 1993, Sarap


and Vashist 1994). There is a greater likelihood that credit constraints may createobstacles for the expansion of HYV area and the use of an optimum dose of inputs.Since credit is given by the lender on the basis of collateral provided by the borrower(Eswaran and Kotwal 1986, Sarap 1990), a poor farmer may not be in a position toraise a sufficient amount of working capital for the complete adoption of new tech-nology. Thus, despite the availability of infrastructure, a lack of investible cash mayretard the adoption of MVs and their component technologies.

TenureCertain forms of tenurial arrangements may also affect the adoption of MVs.Parthasarathy and Prasad (1978) observed that tenants had a lower tendency to adoptHYVs than owners did. In contrast, some empirical studies do not find a clear rela-tionship between tenure and adoption. Vyas (1975) cites studies referring to HYVwheat adoption in India that show that tenants were not only as innovative as land-owners but sometimes used more fertilizer per hectare than owners did. It has beenpointed out that a distinction should be drawn between pure tenants and tenant-own-ers because the latter can be expected to be more receptive to innovations. Schutjerand Van der Veen (1977) have opined that any observed changes were possibly due tothe discriminatory access of the tenants to credit, input markets, product markets, andtechnical information. If these variables vary in different sociocultural environments,empirical results may seem to be in conflict if the underlying factors are not consid-ered directly. In a backward agricultural area, however, where the landlord is an im-portant source of finance, the landlord may discourage the tenant from adopting theMVs and associated technologies or provide inputs at his own convenience lest thetenants’ dependence on the landlord decrease. As such, tenancy may be negativelyassociated with the adoption of MVs.

Supply constraintsThe unavailability of complementary inputs is an important constraint to the adoptionof MVs. In most cases, the high yield potential of the seed can be realized only if atleast some fertilizers are applied (Feder et al 1985). Other inputs are also complemen-tary to different degrees. Several studies reported the unavailability of modern variet-ies and associated technologies as one of the major constraints to the adoption or slowadoption of the same (Behra and Sahoo 1975, Herdt and Wickham 1978, Gupta andKurkute 1978, Sarup and Pandey 1982). Satyanarayana and Kiresur (1990) concludedthat the adoption of HYVs was proportional to the availability of complementaryinputs. This issue is more relevant in a state such as Bihar where the input deliverysystem is very weak and input use, including that of fertilizer and credit, is low.

Labor availabilityHYV technology generally requires more labor inputs, so labor shortages may pre-vent adoption. Moreover, new technologies may increase the seasonal demand oflabor, so that adoption is less attractive for those with limited family labor or thoseoperating in areas with less access to labor markets. Hicks and Johnson (1974) have

310 Kumar and Jha

found that a higher rural labor supply leads to greater adoption of labor-intensive ricevarieties in Taiwan. Harris (1972) has found that shortages of family labor explainnonadoption of HYVs in India. Several studies, however, have not found the avail-ability of labor to be a major determinant of the adoption of new technologies (Shakyaand Flinn 1985).

The above list of adoption constraints is not exhaustive and it may differ fromone ecosystem to another. Conflicting conclusions can also emerge from studies fromdifferent regions or countries because of different social, cultural, and institutionalenvironments apart from pure economic factors. It is therefore essential to providecomprehensive information about the interactions among the various factors that gen-erate the observed adoption patterns.

Analysis of constraints

Bihar accounts for nearly 12% of the rice area of the country but its share of total riceproduction is only around 9%. In Bihar, rice occupies around 50% of the gross croppedarea; however, about 70% of the rice area in Bihar is rainfed. The rainfed lowland riceecosystem constitutes the highest area (40%), followed by the rainfed upland riceecosystem (20%) and deepwater rice ecosystem (10%). Because the rainfed lowlandecosystem has the most rice area, this study was designed to identify and analyzeconstraints to the adoption of MVs of rice and their component technologies in thisecosystem.

Data and methodology

The study used both secondary and farm-level primary data. Time-series secondarydata were collected and used to analyze trends in area, production, and yield of rice inBihar from 1970 to 1998 (Table 1). The trends in adoption of HYVs of rice andfertilizer consumption in the state were also analyzed. Compound annual growth rateswere estimated to analyze the trends. To study the decade-wise performance, the spanof 28 years was divided into three periods: (1) 1970-71 to 1979-80, (2) 1980-81 to1989-90, and (3) 1990-91 to 1997-98.

Table 1. Area, production, and yield of rice inBihar.a

Area Production YieldYears (000 ha) (000 t) (kg ha–1)

1970-71 5,134 4,631 9031980-81 5,339 4,496 8381990-91 5,156 5,791 1,1171997-98 5,027 6,899 1,372

aData are for triennium ending average.


Farm-level primary data were obtained from 80 randomly selected farmers,from eight villages belonging to the four development blocks in two each of the north(Samastipur) and south Bihar (Patna) plains. Ten farmers from each of the samplevillages were selected randomly from lists of the farmers having rainfed lowland ricearea. The information on existing farming practices, varieties, yields, input use, dif-ferent adoption constraints, etc., was obtained from the sample farmers with the helpof a comprehensive questionnaire from the selected sample farms. In addition, a com-prehensive list of adoption constraints was given to them and they were asked toassign a value of one to the most limiting adoption constraint, two to the next impor-tant one, and so on. Then the rank values were averaged across the villages and acomposite score was obtained on the basis of which the top ten and five constraintswere identified for the adoption of MVs and fertilizer application, respectively. Thisinformation was cross-checked with the scientists working in the area to make thedata more reliable.


Annual compound growth rates in rice area, production, and yield were computed fordifferent periods in the state (Table 2). It is evident from Table 2 that Bihar did notshow significant growth in rice production and yield during 1970-79. There was aconspicuous change in rice production in the state during the 1980s, however, whenproduction increased at an annual growth rate of 4.05%. Production gained furthermomentum and increased at an annual rate of 5.36% during the 1990s. The increasedrice production in the state during both subperiods (1980-89 and 1990-97) came al-most entirely from yield increments (about 93% and 98%, respectively). Taking allthe periods together (1970-97), a significant growth of 1.65% in yield and 1.38% inproduction was observed. The trend in this chronically stagnant zone with low pro-ductivity was therefore positive.

Area under high-yielding varieties of riceTable 3 shows the coverage of area under HYVs of rice from 1970-71 to 1996-97 inthe state. Analysis reveals that high-yielding rice varieties are gaining wider accept-ability in the state in recent years. Rice area under HYVs was only about 7% of total

Table 2. Growth patterns of area, production, and productiv-ity of rice in Bihar.

Growth rate (%)a

Item1970-79 1980-89 1990-97 1970-97

Area 0.51 0.25 0.09 –0.27*Production 0.29 4.05* 5.36* 1.38**Yield –0.21 3.80** 5.27* 1.65***

aCoefficient of the semi-log regression. ***,**,* = significant at the1%, 5%, and 10% level, respectively.

312 Kumar and Jha

rice area during 1970-71, but increased to 62% during 1997-98. The coverage ofHYV rice area thus increased at a compound annual growth rate of 9%. The adoptionof HYVs of rice in the state during the 1990s (especially after 1992-93) recorded anunprecedented annual compound growth rate of 35%. This transformation was pos-sible due to the joint efforts of ICAR-DOAC in taking the improved technologiesfrom the shelf to the farmers through the compact block frontline demonstration—aunique experiment launched in 1990-91. Since 1990-91, 340 demonstrations havebeen conducted with the active support and cooperation of agricultural universitiesand departments of agriculture. This experiment disseminates knowledge and facili-tates large-scale farmer participation, enabling farmers to choose varieties and a pro-duction package suited to their farm situations (Siddiq 1999).

Fertilizer consumptionTable 4 shows the consumption of chemical fertilizers in Bihar. Fertilizer consump-tion was a mere 10 kg ha–1 in 1970-71, but increased to 17 kg ha–1 in 1980-81 and toa high of 80 kg ha–1 in 1997-98, recording an annual compound growth rate of 7.86%during 1970-97. Fertilizer consumption witnessed a spectacular increase, especiallyin the 1990s (consumption increased by 150% in 7 years). Despite such an impressivejump in the use of chemical fertilizers, the state is still far below the level of fertilizerconsumption in states such as Punjab (178.6 kg ha–1), Haryana (141.6 kg ha–1), Uttar

Table 3. Rice area under HYVs from 1970-71 to 1997-98 inBihar.

Area under Area under Area of rice AnnualYear HYVs rice under HYVs growth ratea

(million ha) (million ha) (%) (%)

1970-71 0.38 5.13 7.4 –1980-81 1.32 5.34 24.7 10.34*1990-91 1.64 5.16 31.8 5.021997-98 3.13 5.03 62.2 35.14*1970-97 – – – 8.98*

a* = significant at 1% level.Source: Fertilizer Statistics (various issues), Fertilizer Association ofIndia, New Delhi, Agriculture, CMIE, September 1999.

Table 4. Fertilizer consumption in Bihar (kg ha–1).

Years N P2O5 K2O NPK

1970-71 7.47 1.66 0.83 9.961980-81 13.47 2.73 1.27 17.471990-91 40.70 11.07 4.40 56.171997-98 62.43 12.77 4.93 80.13

Source: Fertilizer Statistics (various issues), Fertilizer Association ofIndia, New Delhi, Agriculture, CMIE, September 1999.


Pradesh (108 kg ha–1), West Bengal (120 kg ha–1), Tamil Nadu (152 kg ha–1), andAndhra Pradesh (153 kg ha–1). Moreover, the use of fertilizer nutrients was not bal-anced in the state.

Area under traditional and modern varieties of rice on sample farmsTable 5 shows the area under traditional and modern rice varieties on selected farms.An inquiry into the descriptions of land area among the farmers growing rice in therainfed lowland ecosystem in sample districts of the north and south Bihar plainsclearly reveals that rice area in Samastipur District was more than that of Patna Dis-trict. However, the rice area under HYVs was considerably higher in Patna than inSamastipur. This conforms with the district-level secondary data (Jha and Viswanathan1999). The reason for such a big difference in the adoption of MVs between the twodistricts might be the difference in existing irrigation potentials. The actual irrigatedarea in north Bihar is lower and the available sources of irrigation are costly, whichcompels farmers to cultivate traditional varieties.

Adoption of traditional and modern rice varieties on sample farmsTable 6 shows the position of farmer-adopters in both districts. It is clear from thetable that, out of 40 sample farmers in Patna, only two farmers grew MVs exclu-sively, whereas in Samastipur 25 farmers grew only MVs. This shows that improvedmodern varieties are more popular in Patna and MVs were replacing traditional vari-eties (TVs) at a higher rate in this district. The replacement of TVs by MVs wasmeager in Samastipur. However, MVs are gaining wide acceptance along with TVsas 47.5% of the farmers adopted MVs along with TVs. In spite of MVs gaining popu-larity among the farmers, the predominance of TVs was uninterrupted. For adoptersand nonadopters of MVs in the Bihar plains (taking both districts together), the num-ber of farmer-adopters was apparently high. More than 70% of the farmers identifiedwere either cultivating MVs only (33.75%) or MVs and TVs both (37.5%). The per-centage of farmers cultivating only TVs was 28.75%.

Table 5. Area under traditional and modern rice varieties on sample farms.

Area under Area underNo. of Net Area under modern traditional

Sample zones farms cultivated ricea varietiesb varietiesb

area (ha) (ha) (ha)

North Bihar plains 40 86.2 42.3 14.3 28.0(49.1) (33.8) (66.2)

South Bihar plains 40 78.8 32.7 27.4 5.3(41.5) (83.7) (16.3)

Total 80 165.0 75.0 41.7 33.3(45.5) (55.6) (44.4)

aNumbers in parentheses indicate area under rice as a percentage of total cultivable land area.bNumbers in parentheses indicate area under modern and traditional rice varieties as a percent-age of area under rice.

314 Kumar and Jha

Rice varieties in farmers’ fields and recommended varietiesTable 7 lists different varieties that were released for cultivation in Bihar under differ-ent situations of the rainfed lowland ecosystem. It is obvious from Table 7 that thereare only 10 recommended MVs for such a vast lowland area (17.84 million ha) in thestate. Table 6 reveals that, in spite of the presence of MVs, several TVs are popularamong farmers in the study area. Even with MVs, varieties such as Sita and Jaya arebeing cultivated in some areas in shallow rainfed conditions although these varietiesare not suitable for rainfed cultivation. This clearly reflects the poor technical knowl-edge about the varieties among the farmers because of insufficient technical guidanceand poor transfer of technology. This is probably one of the most important reasons

Table 6. Adoption of traditional and modern rice varieties on samplefarms.

No. of No. of No. ofSample Total no. farmers farmers farmersdistrict of farmersa growing MVs growing growing

and TVs both MVs only TVs only

Samastipur 40 19 2 19(north Bihar (100) (47.5) (5.0) (47.5)plains)

Patna 40 11 25 4(south Bihar (100) (27.5) (62.5) (10.0)plains)

Total 80 30 27 23(100) (37.5) (33.8) (28.8)

aNumbers in parentheses indicate percentage of sample farmers in respectivesample zones.

Table 7. Rice varieties recommended for the rainfed lowland ecosys-tem in Bihar.

Land situation Modern varieties Varieties cultivatedreleased by farmers

Shallow lowland Jayshree, Rajshree, Sita, Jaya, BR34,Mahsuri, Vaidehi, Bakol, Ramraji,Radha, Pankaj, Panjabi, Patania,BR3 Permal

Intermediate lowland Jayshree, Radha, Rajshree, BR34,Pankaj, BR8 Radha, Bakol,

Punjabi,Hathijhuln,Chenab, Latisail,Patania, Permal

Semideep Janki, Sudha Janki, Sudha,Hathijhuln, Bakol,Dermi, Pankaj


that create negative impressions about the MVs among the farmers as the injudiciousselection of varieties often results in large failures. This also compels the farmers tocontinue cultivating low-yielding but to some extent reliable TVs.

Yield gap between modern and traditional rice varietiesTable 8 shows the average yield gaps between modern and traditional rice varieties inSamastipur and Patna districts. The average yield of MVs on the sample farms ofSamastipur was 2.6 t ha–1, whereas the average estimated yield of the TVs was 2.2 tha–1. Although the average yield of MVs in Patna was higher (2.7 t ha–1) than that ofSamastipur, the yield of TVs was comparatively lower in Patna (2.1 t ha–1).

Constraints to the adoption of modern varietiesand their component technologiesThe rice area under HYVs and fertilizer use in Bihar have registered a significantincrease in recent years. This progress, however, has not yet reached the expectedlevel and the poor adoption of modern rice varieties under the rainfed lowland eco-system in the Bihar plains is cited as one of the main reasons for the lower productiv-ity of rice in Bihar. Farmers have several constraints to the adoption of MVs and theirassociated technologies.

Table 9 presents different constraints that affect the adoption of modern ricevarieties and that many farmers in the study area often face. The nonavailability ofMVs or unavailability in time has emerged as the most important constraint to theiradoption in the rainfed lowland ecosystem of Bihar. This reaffirms the poor conditionof the state’s input delivery mechanism. Cereal farms in developing countries such asIndia often have three major sources of seed: seed purchased from a formal seedindustry, seed obtained from other farmers, and seed retained from the previous year’sgrain crop (Tetlay et al 1991). More than 85% of the seed consumed in India wasproduced by the farmers themselves (Banerjee 1984). The next important source wasfellow farmers. The share of the organized seed sector was meager on account of thehigh price of certified seeds and nonavailability at proper places in time (Sidhu 1999).The gap between the seed requirement and seed actually distributed in India for paddywas about 52% in 1992-93 (Sidhu 1999). This figure was expected to be higher inBihar. The large gap between the requirement of certified/quality seeds and their dis-

Table 8. Yield gap between modern and tradi-tional rice varieties in the rainfed lowland eco-system in Bihar (kg ha–1). Yield figures are interms of unhusked paddy.

Av yield of Av yield ofSample modern traditional Yield gapdistrict varieties varieties

Samastipur 2,575 2,212 363 (14.1)Patna 2,690 2,056 634 (23.6)

316 Kumar and Jha

tribution in the country is a matter that requires serious attention for increased seedproduction.

The high input cost associated with the cultivation of MVs has also been per-ceived as the next important constraint to MV adoption although a huge amount isbeing spent on input subsidies. The MVs being cultivated in the area are susceptibleto either drought or flood or both and these are also more prone to insect pests anddisease infestation (CRRI 1997). Although partial solutions are available in the formof resistant varieties and diversified cropping patterns, their adoption is often delayedby the slow pace of technology transfer (Siddiq 1996). These factors limit the adop-tion of MVs of rice in the study area considerably. Farmers in this area urgently needa variety that can withstand water stress as well as submergence to some extent. Lowprofitability (relatively) has also emerged as an important constraint. As pointed outby Dr. Norman E. Borlaug, people would adopt technology to increase yield only if itgives returns of 200% or more (Hindu Survey, Agriculture, 1999). He disagrees withthe notion that it is difficult to bring about changes. Such a psychological hurdle canbe removed by the effective demonstration of technology. Poor threshability andthatchability also restrict the adoption of MVs as they limit the use of straw as fodderand for other domestic purposes. Few credit facilities are available to farmers in mostof the eastern states, especially in Assam, Orissa, and Bihar, and they are often lessthan what a farmer in Punjab, Haryana, and Kerala has. Further, the farmers’ access tocredit is frustrated by financial institutions, which adopt complicated procedures forgranting and recovering agricultural loans. A review and necessary improvement ofthe system will help in getting the benefits and facilities provided by the governmentto the farmers in full measure. Bihar also suffers from outdated tenurial relations,which also adversely affect the adoption of MVs in the state. A related major problemis the fragmentation of holdings. There is a need to undertake tenurial reforms in alarge way and simultaneously to consolidate holdings. Institutional development mustalso receive high priority, as it would facilitate the adoption of MVs and their compo-

Table 9. Constraints for low adoption of modern varieties inBihar plains.

CompositeConstraint score Rank

Unavailability of seeds/nonavailability ofseeds in time 1.34 1

High input costs 1.83 2Susceptibility to drought/flood 3.23 3Prone to insect pests and diseases 4.33 4Low profitability 4.88 5Poor threshability and thatchability 5.55 6Unavailability of credit 6.67 7Bad tenurial system 7.11 8Poor extension service 7.60 9Scarcity of labor 7.91 10


nent technologies. Another problem facing this region is the lack of appropriate ex-tension services. Measures have to be devised to acquaint the farmers with modernproduction technology through adequately trained personnel and to provide them withfacilities for soil testing and proper advice on upgrading production technology. More-over, the communication gap between researchers and farmers should be bridged.

Adoption of chemical fertilizerModern rice varieties are highly responsive to chemical fertilizers; however, mostfarmers do not adopt recommended doses of chemical fertilizers in their fields. Table10 presents the factors that affect the application of chemical fertilizers to rice in therainfed lowland production system.

All the farmers responded that high fertilizer costs make their application un-profitable in the rainfed lowland ecosystem because dependence on monsoon andinefficient water management techniques reduce the efficiency of chemical fertiliz-ers. A large portion of chemical fertilizers becomes unavailable to the plant either byleaching or by being fixed in the soil. Besides the high fertilizer cost, the problem ofunavailability of chemical fertilizer during the peak season is one of the most limitingfactors for the application of fertilizer in rainfed lowland rice. Therefore, the timelysupply of good-quality fertilizers, especially in remote areas, is equally important.This requires a revamping of the entire infrastructure for an effective fertilizer supplydistribution system. Farmers also reported that the lack of capital in their commandareas made the purchase of chemical fertilizer economically unaffordable; thus, theyapply the minimum possible doses. Farmers also face difficulty in applying fertilizersbecause their fields were unbunded and thus they believed that fertilizer applicationwould be ineffective. The prevailing adulterated fertilizer in markets also makes farmerswary of its possible adverse effect on the crop.

Conclusions and policy implications

Agriculture in the rainfed lowland ecosystem has performed well in recent years.Production will probably never match that in the most productive irrigated areas be-cause of inferior agroclimatic conditions, but growth potential still exists (Bhalla et al1999). The higher adoption of MVs and other component technologies can become a

Table 10. Factors affecting the adoption of chemical fertiliz-ers on sample farms.

CompositeFactors/constraints score Rank

High fertilizer cost and unprofitable use 1.29 1Unavailability of fertilizers in time 1.78 2Scarcity of capital 2.77 3Operational difficulty 3.33 4Adulteration 3.88 5

318 Kumar and Jha

vehicle for realizing the untapped growth potential in this region. Hence, a twin pillarstrategy of varietal improvement and appropriate improved production technologiesaddressing these constraints would be ideal. The higher adoption of new technologiesin the rainfed rice ecosystem has to be matched with suitable policy initiatives fordesired results. Some critical areas for intervention would be (1) improving the avail-ability of seed/planting material of HYVs, (2) strengthening the input (seed, fertil-izer) delivery system, (3) radically reorienting credit policies and procedures, (4) con-tinuously transferring technology through assessment and refinement as well as im-proving and strengthening the existing extension system, (5) mechanizing small farms,(6) promoting water harvesting for judicious use of water, (7) achieving reforms intenancy and associated laws, (8) extending crop insurance schemes to provide safe-guards against risk and uncertainty, and (9) ensuring remunerative prices and improv-ing marketing infrastructure. Moreover, there is a crucial need for a macro policy forcreating a favorable environment for the better flow of information, investment, in-puts, and technology, although the macro policy in itself is not sufficient to deliver theexpected results. This is largely due to the characteristic features of Indian agricul-ture. This underscores the necessity of decentralized or micro planning if the adop-tion of modern rice varieties and their component technologies is to be raised sub-stantially.

ReferencesAhmad I. 1981. Technical changes and agrarian structure: a study of Bangladesh. Geneva (Swit-

zerland): International Labour Office.Asaduzzaman M. 1979. Adoption of HYV rice in Bangaladesh. Bangladesh Dev. Stud. 7(3):23-

52.Banerjee SK. 1984. Seed in Indian agriculture during 1980s and beyond. Indian Agric. 28(2):

115-120.Barker R, Herdt RW. 1978. Equity implications of technology changes. In: Interpretive analy-

sis of selected papers from Changes in Rice Farming in Selected Areas of Asia. LosBaños (Philippines): International Rice Research Institute.

Behra C, Sahoo MS. 1975. Impact of national demonstration on adoption of agricultural prac-tices. Indian J. Ext. Educ. 11(1-2):32.

Bhalla GS, Hazell P, Kerr J. 1999. Prospects for India’s cereal supply and demand to 2020.Food, Agriculture, and the Environment Discussion Paper 29. Washington, D.C. (USA):International Food Policy Research Institute.

Bhalla Surjit S. 1979. Farm and technical changes in Indian agriculture. In: Berry R, Cline W,editors. Agrarian structure and productivity in developing countries. Baltimore, Md.(USA): Johns Hopkins University Press.

Bhati UN. 1976. Some social and economic aspects of the introduction of new varieties ofpaddy in Malaysia: village case study. Geneva (Switzerland): United Nations.

CRRI (Central Rice Research Institute). 1997. Perspective plan. Cuttack (India): CRRI.Eswaran M, Kotwal A. 1986. Access to capital and agrarian production organization. Econ. J.

96(382):482-498.


Feder G, Just RE, Zilberman D. 1985. Adoption of agricultural innovations in developing coun-tries: a survey. Econ. Dev. Cult. Change 33(2):263-283.

Frankel FR. 1971. India’s green revolution: economic gains and political costs. Princeton, N.J.(USA): Princeton University Press.

Fuglie KO. 1992. The adoption of new agricultural technology in a rainfed rice farming systemin northeast Thailand. Unpublished thesis, Humanities and Social Sciences, Universityof Minnesota, Minneapolis, Minn., USA.

Gowda Channe MB, Jolihal KA. 1969. Adoption of recommended varieties and seedling prac-tices of paddy by farmers. Indian J. Ext. Educ. 10(1&2):70.

Green DAG, Ng’ong’ola DH. 1993. Factors affecting fertilizer adoption in less developedcountries: an application of multivariate logistic analysis in Malawi. J. Agric. Econ.44(1):99-110.

Gupta DS, Kurkute KD. 1978. New rice technology and small farmers. Eastern Econ. 71(14):666-668.

Harris B. 1972. Innovation adoption in Indian agriculture: the high yielding variety program.Modern Asian Stud. 6.

Hayami V. 1981. Induced innovation, green revolution, and income distribution: comment.Econ. Dev. Cult. Change 30:169-176.

Hazell PBR. 1982. Instability in Indian foodgrain production. Research Report 30. Washing-ton, D.C. (USA): International Food Policy Research Institute.

Herdt RW, Wickham TH. 1978. Exploring the gap between potential and actual rice yields: thePhilippine case. In: Economic consequences of the new rice technology. Los Baños,Laguna (Philippines): International Rice Research Institute. p 3-29.

Hicks WH, Johnson R. 1974. Population growth and the adoption of new technology in Tai-wanese agriculture. Working Paper in Economics No. 1974-E6. Columbia, Mo. (USA):University of Missouri.

Jha TN, Viswanathan. 1999. Problems and prospects of agricultural development in Bihar.Occasional Paper-10. Mumbai (India): National Bank for Agriculture and Rural Devel-opment.

Khan MH. 1975. The economics of the Green Revolution in Pakistan. New York (USA):Frederick A.

Misra B, Mukhopadhyay SK, Flinn JC. 1986. Production constraints of rainfed lowland rice ineastern India. In: Progress in rainfed lowland rice. Los Baños, Laguna (Philippines):International Rice Research Institute. p 25-52.

Mitra A. 1971. Delivery and distribution of inputs of agriculture to small farmers in India. ABackground Paper for the Raman Magsaysay Award Foundation Seminar held at Manila.

Muthia C. 1971. The Green Revolution: participation by small versus large farmers. Indian J.Agric. Econ. 26:25-34.

Palmer I. 1976. The new rice in Asia: conclusion from four country studies: Report of theUnited Nations Research Institute of Social Development, Geneva.

Pandey S, Singh HN, Villano RA. 2000. Rainfed rice and risk-coping strategies: somemicroeconomic evidence from eastern Uttar Pradesh. In: Pandey S, Barah BC, VillanoRA, Pal S, editors. Risk analysis and and management in rainfed rice systems. LimitedProceedings of the NCAP/IRRI Workshop on Risk Analysis and Management in RainfedRice Systems, 21-23 Sep 1998, National Centre for Agricultural Economics and PolicyResearch, New Delhi, India. Los Baños (Philippines): International Rice Research Insti-tute. p 115-134.

320 Kumar and Jha

Parthasarathy G, Prasad DS. 1978. Response to the impact of the new rice technology by farmsize and tenure: Andhra Pradesh. Los Baños (Philippines): International Rice ResearchInstitute.

Rehman RI. 1983. Adoption of HYV: role of availability and the supply side problems.Bangladesh Dev. Stud. 11(4).

Roy GL. 1976. Reasons for low spread of HYV paddy during the pre-kharif and kharif seasonin West Bengal. Indian J. Ext. Educ. 12(1-2):50.

Sarap K. 1990. Factors affecting small farmers’ access to institutional credit in rural Orissa,India Dev. Changes 21(2):281-307.

Sarap K, Vashist DC. 1994. Adoption of modern varieties of rice in Orissa: a farm level analy-sis. Indian J. Agric. Econ. 49(1):88-93.

Sarup S, Pandey RK. 1982. Investigation into non-adoption of rice varieties in Orissa: an appli-cation of discriminant function techniques. Margin 15(1):71-77.

Satyanarayana M, Kiresur VR. 1990. An investigation into partial adoption of HYVs of kharifrice in India. Agric. Situation India 45(5):339-344.

Schluter M. 1971. Differential rates of adoption of the new seed varieties in India: the problemof small farms. U.S. Agency for International Development, Occasional Paper No. 47.Ithaca, N.Y. (USA): Cornell University.

Schutjer W, Van der Veen M. 1977. Economic constraints on agricultural technology adoptionin developing countries. U.S. Agency for International Development, Occasional PaperNo. 5. Washington, D.C. (USA): USAID.

Shakya PB, Flinn JC. 1985. Adoption of modern varieties and fertilizer use on rice in theeastern Tarai of Nepal. J. Agric. Econ. 36(3):409-419.

Sharma AC. 1973. Influence of certain economic and technological factors on the distributionof cropped area under various crops in the Ludhiana District. J. Res. Punjab Agric. Univ.10:243-249.

Siddiq EA. 1996. Rice: development opportunities. Hindu Surv. Agric.Siddiq EA. 1999. Rice: not a distant dream. Hindu Surv. Agric.Sidhu MS. 1999. Impact of intellectual property rights on the Indian seed industry. Indian J.

Agric. Econ. 54(3):370-379.Singh AJ, Byerlee D. 1990. Relative variability in wheat yields across countries and over time.

J. Agric. Econ. 41:21-32.Tetlay KA, Heisey PW, Ahmed Z, Munir A. 1991. Farmers’ sources of wheat seed and wheat

seed management in three irrigated regions of Pakistan. Seed Sci. Technol. 19(1):123-138.

Thorner D. 1964. Agricultural cooperative in India: a field report. Bombay (India): Asia Pub-lishing House.

Tripathi A. 1977. A study of technology gap in adoption of new rice technology in coastalOrissa and constraints responsible for the same. Unpublished PhD thesis. Division ofAgricultural Extension, Indian Agricultural Research Institute, New Delhi.

von Pischke J. 1978. When is smallholder credit necessary? Dev. Dig. 26:6-14.Vyas VS. 1975. India’s high yielding varieties programme in wheat. 1966/7-1971/2. Mexico

City (Mexico): Centro Internacional de Mejoramiento de Maíz y Trigo.Walker TS, Ryan JG. 1990. Village and household economies in India’s semi-arid tropics.

Baltimore, Md. (USA): Johns Hopkins University Press.Wills IR. 1972. Projection of effects of modern inputs on agricultural income and employment

in a C.D. Block, U.P., India. Am. J. Agric. Econ. 54(3):452-460.


NotesAuthors’ address: National Centre for Agricultural Economics and Policy Research, Pusa, New

Delhi-12, India.Citation: Tuong TP, Kam SP, Wade L, Pandey S, Bouman BAM, Hardy B, editors. 2000. Char-


322 Kumar and Jha

Appendix. Agroclimatic zones of BiharThe state of Bihar is divided into six agroclimatic zones based on rainfall and temperature, soiltype, and physiographic features: (I) northwest alluvial plains, (2) northeast alluvial plains, (3)south Bihar alluvial plains, (4) central and northeastern plateau, (5) western plateau, and (6)southeastern plateau. Of these, the first three form the segment of the middle Gangetic plains.The climate is dry to moist subhumid and the soil type is heavy-textured sandy loam to clayey,medium acidic. This subzone receives more than 1,200 mm of rainfall annually. The last threeform part of the eastern plateau and hills. The climate of this subzone is moist subhumid tosubhumid and the soil is red sandy, loamy, red, and yellow. It receives a higher rainfall ofaround 1,300–1,400 mm but irrigation development is very poor as only 8% to 10% of thecultivated area is irrigated.

The Rajendra Agricultural University, Pusa, however, classified Bihar into four compre-hensive agroclimatic zones: (1) northwest alluvial plain, (2) east alluvial plain, (3) southwestalluvial plain, and (4) plateau region.

The selected districts, Samastipur and Patna, come under the northwest alluvial plainand southwest alluvial plain, respectively. The southwest alluvial plain has about 33% of thestate’s total production of rice with around 24% area. The northwest plains contribute about26% to rice production in the state. The east plains contribute about 18%. The rest is contrib-uted by the plateau region. Across different zones, the southwest zone displayed higher yield(1.58 t ha–1) than the northwest plains (1.08 t ha–1).

Rainfed rice, risk, and technology adoption: . . . 323

Risk and risk aversion are often believed to be important factors that con-strain the adoption of technology by poor farmers. The nature and magnitudeof risk and how farmers manage risk, however, have been poorly studied inthe context of rainfed rice. This chapter provides an analysis of variability inrice production and farmers’ coping strategies using farm-level panel datafrom two villages of eastern Uttar Pradesh, India.

The results indicate that crop diversification is an important farmer strat-egy for dealing with risk. Through crop and income diversification, farmershave been able to reduce the effect of risk in rice production on the variabilityof total household income. The economic cost of risk in rice production wasfound to be quite low, implying that yield-increasing rather than yield-stabiliz-ing rice technologies are likely to be more appropriate in these environments.The adoption of modern varieties was conditional on irrigation, farmers’ edu-cation, farmers’ age, and their wealth status as represented by farm size.Implications of these results for technology design and policy reforms arediscussed.

Uncertainties involved in rainfed rice farming are manifold and farmers face consid-erable challenges in dealing with these. Of all the elements of risk, the most importantfor agriculture is rainfall. Risk under rainfed conditions generally tends to be high asvariability in rainfall can lead to wide swings in yield and output. Poor farmers, whenfaced with high levels of risk, may respond by adopting practices that reduce riskeven if they entail a reduction in income on average. To the extent that risk and riskaversion constrain technology adoption, the study of risk and farmers’ mechanismsfor coping with risk is important for technology design and policy reforms.

Farmers cope with risk by developing various strategies over time. These strat-egies can be classified into ex ante and ex post depending on whether they help re-duce risk during the production process or reduce the impact of risk after a productionshortfall has occurred. Because of the absence of efficient market-based mechanismsfor diffusing risk, farmers modify their production practices to provide self-insurance

Rainfed rice, risk, and technologyadoption: microeconomic evidencefrom eastern IndiaH.N. Singh, S. Pandey, and R.A. Villano

324 Singh et al

such that losses are reduced during unfavorable years (Pandey et al 1998). Ex antestrategies can be broadly grouped into two categories: those that reduce risk by diver-sification and those that do so by maintaining flexibility. Spatial diversification offarms, diversification of agricultural enterprises, diversification of crops and variet-ies, and diversification of income among farm and nonfarm activities are some ex-amples (Siddiq and Kundu 1993, Singh et al 1995). Similarly, maintaining flexibilityis an adaptive strategy that allows farmers to switch resources among activities asrequired to reduce risk.

Loss management or ex post strategies are designed to prevent a shortfall inconsumption when the family income drops below what is necessary for maintainingconsumption at its normal level. These include seasonal migration, consumption loans,and asset liquidation (Jodha 1978). However, these strategies are often inadequate toprevent a reduction in consumption (Pandey et al 2000).

The objective in this chapter is to characterize the nature and magnitude of riskin rainfed rice areas of eastern Uttar Pradesh, India. Farmers’ ex ante risk-copingstrategies and the extent to which risk and risk aversion may be constraints to tech-nology adoption are also analyzed.

Methodology and data

The study of risk and risk management strategies requires temporal data as it is theuncertainty in production and income over time that is of concern to individual farm-ers. Such data also permit an assessment of the dynamic nature of risk and adjust-ments that farm households undertake in response to risk. Accordingly, the study isbased on panel data collected from two villages in eastern Uttar Pradesh for the pe-riod 1994-98. Ninety respondent farmers were randomly selected, 45 farmers fromeach village, and detailed data on rice production practices and other farm and non-farm activities of farm households were collected using the interview method. Pro-duction data were collected at the plot level.

The two villages were selected to provide a contrast in production systems. Thevillage Mungeshpur has better access to irrigation, has more lowland fields, and theadoption of modern varieties is greater. In contrast, Itgaon has a poorer irrigationinfrastructure, has a relatively higher proportion of upland fields that are more easilydrained, and the adoption of modern varieties is lower (Table 1). Thus, Mungeshpurprovides a “benchmark” against which the production strategies of farmers in themore risk-prone Itgaon could be assessed. A comparison of rice production practicesin these two villages provided the basis for assessing farmers’ responses to differen-tial risks.

Our analytical method consisted of computing the variability of rice yield, in-put use, and income at both the plot and farm level. Data were analyzed by poolingthe cross-sectional data with temporal data. To remove farmer-specific effects in plot-level data, a fixed-effects model was specified using farmer-specific dummy vari-ables. The estimating equation was of the form


Yijt = a + bFj + eijt (1)

where Yijt is the yield of the ith field for the jth farmer in year t, Fj is the dummyvariable for the jth farmer, and e is the random error term. The residuals of this dummyvariable model were used to estimate the variance of yield and net returns from rice.The availability of panel data also permitted the estimation of variances of rice in-come and total household income for each household over time. Estimates of thesetemporal variances and coefficients of variation were then used to assess the potentialeffect of uncertainty in rice production on the variability of household incomes.

Crop diversification is a potentially risk-reducing strategy. If crop yields acrossdifferent crops are poorly or negatively correlated, crop diversification can reduce theinstability in total output. The extent of crop diversification was computed by anindex using equation (2):

D = 1 – Σ (aij /Ai)2 (2)j

where aij is the area planted to the jth crop by the ith farmer and Ai is the total croppedarea of the ith farmer. The index D ranges from zero for complete specialization inone crop to the maximum value of unity indicating a high degree of diversification.

If farmers choose activities that generate a lower average income over time inan attempt to reduce risk, the “cost of risk” is the mean income forgone. The propor-tional risk premium is the cost of risk expressed as a proportion of the total meanincome. Under the simplifying assumption that the net returns are normally distrib-uted, the proportional risk premium can be approximated by the following equation(Pandey et al 1999):

Table 1. Basic characteristics of the study area.

Characteristics Mungeshpur Itgaon

No. of respondents 45 45Irrigated area (%) 85 37Average operational 0.7 1.4

holding (ha)Proportion of land type (%)

Upland 34 55Medium land 14 24Lowland 52 21

Average years of 5 6schooling ofhousehold head

Average household 7 8size (no. of members)

326 Singh et al

P = 0.5 R [a2 Cr2 + 2 a (1 – a) g Cr Cy] (3)

where R is the coefficient of relative risk aversion, Cr is the coefficient of variation(CV) of rice income, a is the share of rice income in total income, Cy is the CV ofnonrice income, and g is the correlation coefficient between the rice and nonriceincome. The proportional risk premium measured in equation (3) provides an esti-mate of the cost of risk currently borne by farmers relative to the situation in whichthe variability of rice income is completely eliminated. As there will always be somevariability of rice income that cannot be eliminated, the estimate obtained from equa-tion (3) can be considered to be an upper bound value.

In addition, the determinants of adoption of modern varieties were identifiedusing the probit model (Greene 1997). The factors determining the adoption of mod-ern varieties on individual plots were identified using plot-level data. Regressors usedfor this analysis were a set of household-specific factors and a set of plot-specificfactors. The probit coefficients were used to calculate the marginal probabilities ofadoption associated with various regressor variables. The use of the probit modelpermitted the simultaneous consideration of the effects of biophysical and socioeco-nomic factors in the adoption of modern varieties1.

Results

Table 2 shows summary statistics regarding rainfall from a location near the studyvillages. These data indicate that rainfall variability is high at both the initial stage ofland preparation/rice establishment (June/July) and during the later stage of crop growth(October). Compared with 1994 and 1997, rainfall in June/July was much lower in1995 and 1996. Of the two drought years, 1995 and 1996, the early season drought in1995 was more severe than in 1996. Based on the distribution of and the amount ofrainfall, 1994 may be classified as a “normal” year, 1995 and 1996 as “drought”years, with drought being more severe in 1995, and 1997 and 1998 as “good” years.The surveyed years thus provide a range of rainfall scenarios for investigating farm-ers’ responses.

1As panel data are used to estimate the probit model, a brief digression on the econometric issues involved isin order. Such data permit the specification of farmer-specific effects, which can be treated as fixed or randomeffects (Greene 1997). A simple way of incorporating fixed effects is through the use of farmer-specific dummyvariables. However, with a large number of cross-sectional units (60 in the present data set), the estimates ofindividual fixed effects are likely to be somewhat intractable (Greene 1997, p. 897). Estimation of the randomeffects model is a possibility but at the added cost of computational burden. Instead of these more sophisti-cated methods, we have treated the panel data as a single cross section by ignoring the farmer-specific effects.While such a simplification is likely to lead to parameter estimates that are statistically less efficient, any biason the estimated slope coefficients is likely to be minimal due to the low correlation of the independent vari-ables used with the excluded farmer-specific dummy variables. Nevertheless, this caveat should be kept inmind while interpreting the coefficients of the probit model estimated.


Variability of rice yield and net returnsThe average rice yield in Mungeshpur is about 0.5 t ha–1 higher than in Itgaon (Table3). The variability of yield, as measured by the coefficient of variation, however, islower in Mungeshpur. A higher mean and a lower CV are indications of a more favor-able environmental condition for rice production in Mungeshpur. For both Mungeshpur

Table 2. Kharif-season rainfall for Kumarganj, Faizabad, in 1994-98.

June July August September October Total

Year No. of Rain No. of Rain No. of Rain No. of Rain No. of Rain No. of Rainrainy (mm) rainy (mm) rainy (mm) rainy (mm) rainy (mm) rainy (mm)days days days days days days

1994 8 145 15 224 16 341 6 153 1 2 46 8651995 8 109 7 60 23 355 10 395 0 0 48 9191996 11 99 5 133 18 337 14 168 3 165 51 9021997 4 121 19 485 15 264 12 294 2 43 52 1,2071998 4 29 19 337 16 276 6 110 2 47 47 799

CRS, Masodha, Faizabad (1967-98)Long-term 145 314 300 141 74 1,088

mean rainfall (mm)Coefficient 87 45 40 56 115 22

of variation of long-term rainfall (%)

Table 3. Measures of probability distribution ofplot-level rice yield, net returns, and gross re-turns in rice (1994-98).a

Item Mungeshpur Itgaon

YieldMean (t ha–1) 2.2 1.7CV (%) 52 60Skewness 2.88 0.79Gross returnsMean (US$ ha–1) 210 140CV (%) 62 63Skewness 4.67 0.94Net returnsMean (US$ ha–1) 177 112CV (%) 73 84Skewness 3.72 0.77

aAll monetary values are at 1994 constant prices. Thecoefficient of variation (CV) of yield was calculated us-ing the fixed-effects model specified in equation (1).Values in local currency were converted using US$1 =Rs 31.38.

328 Singh et al

Table 4. Crop diversification index.

Mungeshpur ItgaonYear

Kharif Rabi Kharif Rabi

1994 0.47 0.56 0.72 0.711995 0.53 0.55 0.90 0.771996 0.58 0.57 0.82 0.671997 0.50 0.53 0.75 0.681998 0.45 0.53 0.78 0.62

Table 5. Coefficient of variation (CV) of income from different sources,1994-98.a


CV of rice income (%) 30 76CV of all rainy-season crop income (%) 28 23CV of all crop income (%) 10 9CV of all income 9 12

aThese coefficients of variation are the average values (across households) ofthe coefficient of variation for each household calculated using temporal datafor that household.

and Itgaon, however, the plot-level CVs of yield are quite high by conventional stan-dards. The CVs of net returns (defined as gross returns minus cash costs) are evenhigher and are indicative of the high levels of variability that farmers in rainfed envi-ronments have to deal with. The large difference between the CV of yield and netreturns also indicates that yield variability can severely underestimate risk if the farm-ers’ objective is to obtain stable net returns. Analysis based on yield variability alonecan hence be misleading.

Crop diversification and riskThe diversification indices (Table 4) indicate that the cropping pattern in Itgaon ismore diversified than in Mungeshpur for both the kharif (rainy) and rabi (postrainy)seasons. In Itgaon, farmers grow rice in proportionately smaller areas than inMungeshpur and rely more on pulses and maize, which grow well under rainfed con-ditions. Mixed cropping and intercropping are also more common in Itgaon. Of thefive years, the kharif season diversification indices are higher during the droughtyears (1995 and 1996). These results indicate that crop diversification may be animportant risk-reducing strategy, especially to farmers of Itgaon. The average CV ofincome from all rainy-season crops is substantially lower than the CV of rice income(Table 5). Using a multiple regression analysis of the CV of rainy-season crop incomeon farm size and diversification index, Pandey et al (1999) show that more diversifiedfarms in Itgaon do have a lower variability of crop income. The stabilization effect is


even stronger if the CV of income from all crops (both the kharif and rabi seasons) isconsidered (Table 5).

The variability of plot-level net returns from rice reported in Table 3 was foundto be higher than the variability of total net returns from all plots of rice cultivated bya household (Table 5). Low correlation among plot-level net returns may have led toa reduction in CV of total net returns when rice is grown in several fields with varyingsoil and hydrological characteristics. Farmers’ land portfolios that include these vary-ing field characteristics may thus help stabilize income even though there are likely tobe some efficiency losses in managing different field types that are often spatiallyscattered.

Importance of rice in farmers’ incomeTable 6 shows the relative share of income from rice (i.e., the value of rice producedminus the cash costs) as a proportion of the total household income. It is interesting toobserve that rice accounts for only a small proportion of total household income,although the share of rice increases as farm size increases in Mungeshpur. Farmers’incomes are diversified. Rabi crops and nonfarm income contribute 30% and 46%,

Table 6. Percentage share of different sourcesof income in total household income, 1994-98.

Item Mungeshpur Itgaon Bothvillages

Small farmsRice 13 6 11Rabi cropsa 23 19 22Nonfarm 50 65 54Othersb 14 9 13

Medium farmsRice 17 3 10Rabi cropsa 37 27 32Nonfarm 28 63 46Othersb 19 7 12

Large farmsRice 22 6 8Rabi cropsa 37 35 35Nonfarm 13 43 39Othersb 28 16 18

All farmsRice 15 5 9Rabi cropsa 30 30 30Nonfarm 37 52 46Othersb 17 13 15

aRabi crops are mostly wheat intercropped with mus-tard. bIncludes income from other kharif and summercrops, income from livestock, and income as a farm la-borer.

330 Singh et al

respectively, to farmers’ income. Land-use intensity is almost 100% during the rabiseason in both villages with farmers growing several cash crops such as wheat, sugar-cane, vegetables, and pulses. The availability of tubewell irrigation has facilitatedrabi-season cropping in this part of eastern India. Proximity to the town of Faizabadand the city of Lucknow has also led to an expansion of nonfarm income opportuni-ties in these villages. These villages thus do not agree with the stereotype of an east-ern Indian village where rice production is the major economic activity. A diversifiedincome portfolio that is less dependent on rice production has evolved over time as aresult of commercialization of production systems. Naturally, variability in rice pro-duction will have a small effect on the variability of total household income of farm-ers in such systems, even though farmers continue to grow rice for their food security.

Seasonal adjustmentsIn this section, we analyze the responses over time in terms of rice area planted,method of crop establishment, and changes in rice varieties (Table 7). In bothMungeshpur and Itgaon, rice area decreased in 1995 and 1996 in comparison with1994 and 1997, with the reduction being greater in Itgaon. Delayed rainfall in 1995forced many farmers in Itgaon to forego rice completely, whereas farmers inMungeshpur were able to maintain the rice area by using irrigation. Large fluctua-tions in area planted to rice in rainfed environments can be a major source of variabil-ity in rice production unless there are offsetting movements in yields. A variancedecomposition analysis indicated that in Itgaon the variability in area sown accountedfor about half the total production variation (Pandey et al 1998).

Temporal variations in area under each of the crop establishment methods arehigher in Itgaon than in Mungeshpur. Similarly, the proportionate area planted tomodern varieties also shows some temporal variations. Open-ended interviews withthe farmers indicated that they adjust area under modern varieties and crop establish-ment methods from year to year to cope with climatic risk. When rains were insuffi-cient or delayed at the time of land preparation and crop establishment, transplantingwas not possible in fields without supplemental irrigation due to inability to growseedlings and prepare land. In this situation, farmers preferred to establish rice by dryseeding even though the anticipated yields were lower. Similarly, the use of tradi-tional varieties was more common in such years. These temporal adjustments aremanifestations of flexible decision-making through which farmers attempt to reduceproduction and income losses in adverse years.

Table 7. Tactical adjustments to weather uncertainty.

Mungeshpur ItgaonItem

1994 1995 1996 1997 1998 1994 1995 1996 19971998

% area transplanted 78 76 86 88 96 27 39 50 54 68% area under modern varieties 89 93 94 91 97 67 64 54 61 68% area under rice 60 53 54 60 63 33 10 23 36 28% area fallowed 16 20 14 15 14 33 48 33 26 28


Table 8. Cost of instability in rice income.


CVa of rice income (%) 30 76CV of nonrice income (%) 15 15Ratio of rice to total income (%) 15 5Average cost of risk (% of mean income)

if correlation between rice andnonrice income = 0 0.3 0.1

if correlation between rice andnonrice income = 0.2 0.4 0.3

aCV = coefficient of variation.

Cost of riskThe cost of risk was estimated using equation (3) by assuming the value of R to beequal to 2 (Anderson 1995) and using the sample estimates of the CV of income fromrice and nonrice activities. The correlation coefficient between rice and nonrice in-come in these villages that have diversified income sources was assumed to be 0.2.The estimated cost of risk is less than 1% of the mean income (Table 8). In otherwords, risk benefits from the stabilization of income from rice are negligible. Themain reason for the relative unimportance of variability in rice income is the lowshare of rice income in total income. Farmers have already diversified their incomeaway from rice. Thus, stabilization of rice yield and rice income will have little im-pact on the stabilization of total income. Using a simulation model, Pandey et al(1999) report that even a full stabilization of rice income will lead to a reduction inthe CV of total income by only 2% in Itgaon. Naturally, in areas such as the studyvillages, crop production technologies that raise mean income are likely to be morebeneficial than those that reduce the variability of returns only. In areas where theshare of rice income in total income is high, however, the benefits from stabilizationof rice yield are likely to be higher (Fig. 1).

Fig. 1. Relationship between risk premium andshare of rice income. Drawn using the assump-tions in Table 8 and varying the share of rice intotal income.

20

15

10

5

0

Proportional risk premium (%)

0 5 10 15 20 25 30 35 40 45 50

Share of rice in total income (%)

332 Singh et al

Factors determining the adoption of modern varietiesThe extent of adoption of modern varieties in Mungeshpur is quite high (more than90% of area). Although the adoption rate is lower in Itgaon (60% of area), it is never-theless substantial. The major modern varieties grown are Sarjoo-52 and NDR variet-ies (Table 9). Sarjoo-52 was released in 1980 and is a 130-d variety that performs wellwhen supplemental irrigation is available. Its use has spread, however, even to areasunder rainfed conditions and it has replaced Mahsuri, which is of slightly longer du-ration. Over time, Sarjoo-52 has become quite popular and this variety alone nowcovers almost 40% of the rice area in Itgaon. Similarly, another popular modern vari-ety, NDR-97, was released in 1991. Its duration is 95 d and it is grown mostly underupland fields that are proportionately more in Itgaon.

Sarjoo-52 is not only higher yielding but also less risky than other modernvarieties and some traditional varieties. Using a stochastic dominance analysis, Pandeyand Pal (2000) found that Sarjoo-52 outyielded popular traditional varieties in all ofthe sampled years. The cost/returns analysis shown in Table 10 also indicates that,overall, modern varieties currently being grown not only have higher mean net re-turns (defined as the value of output minus the cash cost) but also lower coefficientsof variation relative to traditional varieties. The usual notion that modern varieties aremore risky than the traditional ones is not supported by the data from these studyvillages. The question still remains as to why the adoption rate remains lower inItgaon than in Mungeshpur, and among small farmers compared with large ones.Obviously, factors other than risk must constrain the adoption of these seeminglyless-risky modern varieties.

Table 9. Percentage area under major modernvarieties (MV) at the study sites (1994-98).a

Variety Mungeshpur Itgaon

Sarjoo-52 35 40Mahsuri 20 3NDR-118 7 18Saket-4 6 6Indrasan 6 2NDR-80 5 7Pant-4 4 2NDR-40032 2NDR-359 2Ashwati 2Pant-10 2 3NDR-97 2 16Other NDRs 2Other MVs 6 2

All MVs 100 100

aNumbers have been rounded to nearest whole number.


Several biophysical and socioeconomic factors may condition the adoption ofmodern varieties. A major set of biophysical factors is related to soil fertility and fieldhydrology. Both these factors are controlled to a certain extent by the location of thefield in the toposequence and the soil type. Table 11 presents the results of the probitmodel estimated using plot-level data. The results indicate that farm size, age andeducation of the farmer, and availability of irrigation are the major variables deter-

Table 10. Cost and return analysis of rice by variety, 1994-98.a

Modern TraditionalItem

Mean CV (%) Mean CV (%)

Yield (t ha–1) 2.7 14 1.5 34Cost

Material costs ($ ha–1) 30 53 18 60Labor costs ($ ha–1) 15 59 10 56Total cash cost ($ ha–1) 45 51 27 58

Gross returns ($ ha–1) 264 6 132 18Net returns ($ ha–1) 219 22 105 34

aAll values are at constant 1994 prices. All local currency is converted using the1994 exchange rate of US$1 = Rs 31.38. CV = coefficient of variation.

Table 11. Probit model estimates of modern variety adoption inFaizabad, eastern India (1994-98).

Standard MarginalVariables Coefficients error probabilities

(%)a

Farm size (ha) 0.178* 0.064 3.53Age of household head (years) –0.019* 0.004 –0.38Education of household head 0.038* 0.014 0.75

(years)Nonfarm income (000 Rs) 0.001 0.005 0.02Dummy for uplandb 0.212 0.114 4.20Dummy for irrigationc 0.521* 0.138 10.35Dummy for saline/sodic soilsd –0.042 0.229 –0.84Time trende 0.073* 0.003 1.46Constant 0.745 0.301Sample size 1,057Log-likelihood ratio 109.00**Percent of correct predictions 80

*and ** indicate significance at 5% and 1% levels. aMarginal probability evalu-ated at mean values of continuous independent variables with all dummy vari-ables set equal to unity. b1 if upland and 0 otherwise. c1 if irrigated and 0otherwise. d1 if soil is saline/sodic and 0 otherwise. e1 for 1994, 2 for 1995, 3for 1996, 4 for 1997, and 5 for 1998.

334 Singh et al

mining whether or not modern varieties will be grown in a particular field. The effectof farm size is positive, indicating that, for a given set of conditions, large farmers aremore likely to adopt modern varieties. As discussed by Feder et al (1985), farm sizemay be a proxy for several factors such as wealth status, extent of risk aversion,financial capability to acquire complementary inputs such as fertilizers, and access toinformation and technology. Operators of large farms are better placed in all theseaspects; hence, a greater adoption of modern varieties. An increase in farm size by 1ha raises the probability of adoption of modern varieties by approximately 4%. Al-though land reform to redistribute land may be justified on various grounds, the adop-tion of modern varieties could decline if land redistribution is not accompanied byimproved access to technology and credit by small farmers. Farmer age is negativelycorrelated with adoption, indicating that younger farmers may be more enterprisingand more willing to experiment with modern varieties than older farmers. House-holds with more years of education are more likely to adopt modern varieties. Despitethe statistical significance of these two farmer attributes, the associated marginal prob-abilities are quite small, indicating the limited relevance of these variables for tech-nology targeting. The probability of adoption is 10% higher in irrigated fields than inrainfed fields, other things remaining the same. The positive and significant coeffi-cient of time trend indicates that adoption has increased over time. The coefficient ofthe dummy variable for uplands is statistically significant at the 10% level (but not atthe 5% level), indicating some weak preference of farmers for growing modern vari-eties in upland field types. Soil type did not have a statistically significant effect. Interms of the size of the marginal effect on the adoption of modern varieties, irrigationhas by far the most dominant effect.

Discussion

The results of the study show that the variability of plot-level yield and net returns ofrice in the study villages is high even in Mungeshpur, where rice is now grown mostlyunder irrigated conditions. In Itgaon, where irrigation is more limited, the variabilityof rice yield and area over time is quite high. Although agricultural researchers havefocused their energy on addressing the problem of yield variability, area variability inrainfed environments can often be an important source of production variability. Tech-nologies that help stabilize rice area planted can help reduce variability in rice pro-duction. One of the major causes of area variability is failure in timely crop establish-ment. Varieties that can be established late and crop management technologies thatfacilitate rapid establishment when environmental conditions are most appropriateare needed to reduce the effect of area variability.

One of the major ex ante strategies to deal with risk in rice production is cropdiversification. The coefficient of variation of income from crops grown during therainy season was found to be inversely related to the extent of crop diversification inItgaon. In addition, crop diversification increased during years of low and/or laterains at the beginning of the cropping season. Although further expansion of tubewellirrigation is likely to reduce the importance of crop diversification as a risk-reducing


strategy in the future (Ballabh and Pandey 1999), rice technologies that facilitate cropdiversification will complement farmers’ coping strategies. An example of such tech-nologies may be improvements in direct-seeding methods to reduce the high labordemand associated with transplanting. This will help relax labor constraints that maylimit crop diversification.

Another important risk-coping mechanism of farmers is the maintenance offlexibility in their decision-making processes. The results show that farmers adjustcrop establishment methods and varieties over time depending on the amount of rainreceived at the beginning of the cropping season. Improved varieties that performequally well under direct-seeding or transplanting methods will contribute to flexibil-ity. Similarly, an accurate forecast of the early season rainfall pattern and dissemina-tion of forecasts to farmers will help adapt rice production methods to match rainfall.

High intensity of land use for production of cash crops during the rabi season isan important feature of the study villages. A substantial proportion of household in-come is generated from the rabi crops. In addition, the proximity of study villages tourban centers has expanded the nonfarm employment opportunities available to farmingfamilies in the study villages. As a result, the share of rice income in the total house-hold income in these villages is below 10%. Even though the rice income is highlyunstable, farmers have been able to stabilize total household income through diversi-fication of income away from rice. This strategy seems to have been quite effective instabilizing total household income, even in Itgaon, which has a higher proportion ofrainfed area. Policies such as the development of infrastructure and rural industrial-ization that facilitate income diversification can thus play an important role in reduc-ing the effect of risk in rice production.

A low proportion of rice in total household income also means that risk in riceproduction per se is relatively unimportant in these villages. The estimated cost ofrisk was found to be below 1% of mean income. This implies that breeding programsthat sacrifice some yield gain in the pursuit of yield stability are likely to be lessattractive than those that aim to improve the average yield. Similarly, farmers’ de-mand for crop insurance to stabilize income from rice is likely to remain low. A cau-tionary note is in order here. These conclusions may not be applicable to other partsof eastern India where the share of rice in total income is higher due to the lack ofincome diversification opportunities.

The results of the study also show that the modern varieties currently beinggrown are not necessarily more risky than the traditional varieties. The conventionalwisdom that modern varieties increase risk is not supported by the data. The spread oftubewell irrigation in the study villages contributed to risk reduction by creating morefavorable growing conditions for modern varieties. However, the fact that traditionalvarieties rarely had higher yields than modern varieties over the study period is anindication of a lower risk associated with these more recent modern varieties.

Despite the apparent effectiveness of farmers’ coping mechanisms in reducingrisk associated with rice production, the spread of modern varieties is still constrainedby the unavailability of irrigation and limited farm size. Although access to irrigationhas improved in more recent years because of increasing investments in tubewell

336 Singh et al

irrigation schemes in the study area, further investments are needed to make ground-water more widely available. Access to groundwater among small farmers can also beimproved by encouraging the development of a more effective groundwater market.The development of varieties more suitable to lowland fields can similarly improvethe chance of adoption of modern varieties in these field types.

Concluding remarks

Farmers’ risk-coping mechanisms were reasonably efficient in preventing incomeshortfall in the study region. Crop and income diversification away from rice hasbeen the major strategy of farmers for dealing with risk. Because of the low impor-tance of rice in total household income, risk related to rice production is relativelyunimportant in constraining technology adoption. Accordingly, the trade-off that mayexist between yield gain and stability needs to be carefully considered in designingbreeding programs as farmers in the study villages have been able to reduce riskthrough income diversification.

Although methods and tools for microeconomic analysis of risk are generallyavailable, a lack of temporal farm-level data covering enough periods remains a prob-lem for the analyst. While an ingenious use of cross-sectional data and weather-drivencrop growth models can help in this, temporal data such as the ones collected in thisstudy are essential for conducting a more complete analysis at the farm-householdlevel. Collection of such data would facilitate in-depth analysis of risk and farmers’coping mechanisms.

ReferencesAnderson JR. 1995. Confronting uncertainty in rainfed rice farming: research challenges. In:

Fragile lives in fragile ecosystems. Proceedings of the International Rice Research Con-ference. Los Baños (Philippines): International Rice Research Institute. p 101-108.

Ballabh V, Pandey S. 1999. Transitions in rice production systems in eastern India: evidencesfrom two villages in Uttar Pradesh. Econ. Polit. Wkly. March 27, 1999. p A11-A16.

Feder G, Just RE, Zilberman D. 1985. Adoption of agricultural innovations in developing coun-tries: a survey. Econ. Dev. Cult. Change 33:255-298.

Greene WH. 1997. Econometric analysis. 3rd edition. Princeton, N.J. (USA): Prentice Hall.Jodha NS. 1978. Effectiveness of farmers’ adjustment to risk. Econ. Polit. Wkly. 13(25): A38-

A48.Pandey S, Pal S. 2000. The nature and causes of changes in variability of rice production in

eastern India: a district-level analysis. In: Pandey S, Barah BC, Villano RA, Pal S, edi-tors. Risk analysis and management in rainfed rice systems. Limited Proceedings of theNCAP/IRRI Workshop on Risk Analysis and Management in Rainfed Rice Systems,21-23 September 1998, National Centre for Agricultural Economics and Policy Research,New Delhi, India. Los Baños (Philippines): International Rice Research Institute. p 73-96.

Pandey S, Behura DD, Villano RA, Naik D. 2000. Economic cost of drought and farmers’coping mechanisms: a study of rainfed rice systems in eastern India. Discussion PaperSeries No. 39. Los Baños (Philippines): International Rice Research Institute. 35 p.


Pandey S, Singh HN, Villano RA. 1999. Rainfed rice and risk coping strategies: some micro-economic evidences from eastern India. Selected paper for presentation at the AnnualMeeting of the American Agricultural Economics Association, 8-11 August 1999, Nash-ville, Tennessee, USA.

Pandey S, Singh HN, Villano RA. 1998. Rainfed rice and risk coping strategies: some micro-economic evidences from eastern Uttar Pradesh. Paper presented at the NCAP/IRRIWorkshop on Risk Analysis and Management in Rainfed Rice Systems, 21-23 Septem-ber 1998, New Delhi, India.

Siddiq E, Kundu DK. 1993. Production strategies for rice-based cropping systems in the humidtropics. In: Buxton DR et al, editors. International Crop Science 1. Madison, Wis. (USA):Crop Science Society of America.

Singh HN, Singh JN, Singh RK. 1995. Risk management by rainfed lowland rice farmers ineastern India. In: Fragile lives in fragile environments. Proceedings of the InternationalRice Research Conference. Los Baños (Philippines): International Rice Research Insti-tute. p 135-148.

NotesAuthors’ addresses: H.N. Singh, Narendra Dev University of Agriculture and Technology,

Faizabad, India; S. Pandey, R.A. Villano, Social Sciences Division, International RiceResearch Institute, DAPO Box 7777, Metro Manila, Philippines.


Using gender analysis in characterizing and understanding . . . 339

This chapter uses gender analysis in characterizing and understanding farm-household systems in typical rainfed lowland rice villages in Faizabad districtin eastern Uttar Pradesh, India. It uses different methods of data collectionsuch as household surveys (structured and unstructured interviews), theparticipatory rural appraisal, and focused group interviews. Results of thestudy reveal that small farming households from the lower caste tend toexploit their female family members to meet competing labor demands be-tween farm and home-based activities. Women from the lower caste provide60% to 80% of the total labor input in rice production. They participate inalmost all rice operations, except in land preparation and application of chemi-cals. When valued, the labor contributions of female members on their ownfarms and through exchange arrangements make up about 20% of the totallabor costs in rice production per hectare. Women’s labor is also crucial tononrice crops and livestock, which are integral in rainfed rice farming sys-tems. Because of the significant contributions of poor women in farming,their roles and needs should be considered in technology development anddissemination. Efforts are now being made to provide women farmers withaccess to new information and new seeds by involving them in the earlyevaluation of rice genotypes through participatory rice varietal selection indrought and submergence rice environments in eastern India.

In recent decades, greater attention has been given to rice research on rainfed low-lands, which cover 48 million hectares in the humid and subhumid tropics of Southand Southeast Asia. Farmers in these ecosystems face adverse biophysical, socioeco-nomic, and cultural constraints to increasing rice productivity. Because of the season-ally variable and erratic rainfall pattern, heterogeneous land types, and diverse socio-economic groups with limited resources, average rice yields are approximately2.3 t ha–1 and are as low as 1.3 t ha–1 (IRRI 1997). Risk under rainfed conditionsusually tends to be high as variability in rainfall can lead to wide swings in yield andoutput (Pandey et al 1998). Because of this uncertainty, farmers do not have the in-

Using gender analysis in characterizingand understanding farm-householdsystems in rainfed lowland riceenvironmentsT. Paris, A. Singh, M. Hossain, and J. Luis

340 Paris et al

centive to invest in cash inputs and devote more time to crop management. Poormanagement has led to low yields, no or low marketable surplus, and consequentlylow income from rice. Farmers are thus caught in a vicious circle of poverty. Despitethe riskiness in rice production, farmers grow rice as the major crop to sustain house-hold food security. Their primary concern is to meet their basic food requirementsand fodder needs for their livestock through their own production. To reduce risk,they diversify their income sources, one of which is for male members to migrate toother cities or other highly productive agricultural areas. This requires allocation offamily labor to various livelihood activities according to the gender roles prescribedby the household and community, degrees of labor specialization by family members,and opportunity costs of family labor.

While there has been greater awareness and recognition of the vital roles thatpoor women play in rice-based farming systems, their unpaid labor contribution inrice farming is seldom valued. This has often led to their exclusion as cooperators inon-farm research and as recipients of training and extension programs. With the in-creasing male migration (seasonal or semipermanent), other family members, par-ticularly the female members, are left behind to manage the farm (crops and live-stock) aside from their daily household and child-care responsibilities. The changinggender roles and responsibilities will have far-reaching implications not only for cropproduction but also for the social organization aspects of the rice household economy.It is important for biological and social scientists to understand the emerging changesthat will shape the nature of rice production systems and their implications for poorhouseholds and gender relations. This important understanding will help prioritizeresearch issues for technology and policy interventions that can improve the well-being of members of farming households, especially the women.

Thus, an analysis of gender roles and relations should be integral when charac-terizing target recommendation domains for technology development. This chapterdiscusses the objectives of gender analysis in on-farm research, and provides a con-ceptual framework for the livelihood systems. It also presents a case in the rainfedlowland rice ecosystems in eastern Uttar Pradesh, India, which demonstrates howgender analysis can be used in characterizing and understanding the farm householdand its environment.

Gender analysis and its objectives

“Gender” refers to a social rather than biological construct, whereas “sex” refers tothe biological differences between men and women. Gender describes the sociallydetermined attributes of men and women, including male and female roles. As a so-cial construct, gender roles are based on learned behavior as a response to socioeco-nomic and environmental pressures and conditions and are flexible and variable acrossand within cultures. Gender is a useful socioeconomic variable to analyze roles, re-sponsibilities, constraints, opportunities, and incentives of the people involved in re-search and development efforts (Poats 1990). It is relational in focus, that is, it isconcerned with women and men in relation to each other. Gender analysis is an ana-


lytical tool used to identify or distinguish the actual “doers” of tasks, decision makersand the potential users of technologies. A primary social mechanism by which menand women are bound into a relation of interdependence is the gender division oflabor (GDOL) in different tasks in farm activities. This will answer the questions

● Who does specific crop and livestock operations?● Who influences a particular activity to be improved, changed, or eliminated?● Whose resources (e.g., labor) will be displaced/increased with the proposed

change?● Who has the incentive to accept the technology or will directly benefit from

it?Knowing the user and beneficiary has both equity and efficiency implications.

This increases the efficiency of farming systems research through targeting and speci-fication, and takes into account patterns and activities of resource use (Feldstein et al1989). An improved understanding of gender roles means that women who were of-ten overlooked will be recognized while their needs, constraints, and productive op-portunities can be addressed by agricultural research and extension. Gender analysisenables scientists to target women’s needs better and predict the impact of plannedinterventions on women.

The conceptual framework for farm-household and livelihood systems

Through gender analysis, women’s roles are not seen in isolation; rather, their rolesand responsibilities in relation to men are analyzed within the context of the complexinteraction between the farm household and the environment (biophysical, social,economic, and cultural). The complexity of rainfed lowland rice production systemscan be best understood by using a holistic conceptual framework for the livelihoodsystem (adapted from Norman et al 1983). A livelihood system includes the farmingsystem and off-farm and on-farm linkages. The farming system consists of a complexinteraction of several independent components (Fig. 1). These components can bedivided into two elements: technical and human. The technical element determinesthe type and physical potential of crop and livestock enterprises, and includes thephysical and biological factors that can be modified through technology develop-ment. For example, the crop calendar could be adjusted to avoid drought or flood bygrowing early maturing varieties or by changing the crop establishment method fromtransplanting to direct seeding. Consequently, these changes will have a major effecton gender roles in rice farming.

The human element is characterized by exogenous (community structure, ex-ternal institutions, etc.) and endogenous factors that can be controlled by the house-holds. At the center of this interaction is the farm household, which is often treated asa “black box” or homogeneous unit represented by the male head of the householdwith members having shared and equal access to resources and benefits from produc-tion. This assumption ignores the differing roles and sometimes conflicting interestsin resource use within a household that affect technology adoption or that in turn cannegatively affect a specific user group (Feldstein et al 1989). There is often inequity

342 Paris et al

Fig. 1. Conceptual framework for understanding the interrelationship between the environmentand farm-household and gender relations in rainfed lowland rice-based farming systems.

in access to and control of resources between male and female members within thehousehold and their communities as a result of norms and traditions that dictate theappropriate behavior and activities of men and women. Studies show that there is anunequal allocation of resources within the households, and that adult female mem-bers and daughters are the most disadvantaged, especially in South Asia (Agarwal1998).

It is also assumed that household labor is homogeneous and thus is freely sub-stitutable across all household tasks, from household to off-farm wage employment.

Physical and biological conditionsSocioeconomic and

cultural circumstances

Physical– Climate– Rainfall– Occurrence

of droughtor flood

– Land type

Biological– Pests– Diseases– Weeds

Soil and topographySoil typeSlope

● Community structures,norms, beliefs

● Kinship (nuclearextended)

● Caste system● Land tenure● Infrastructure● Markets● Extension services● Wages● Prices

Male and femalehousehold members

Resource allocation

Land Water Capital Labor Human capital– Education– Technical

knowledge

Livelihood system

Croppingsystem

Livestocksystem

Off-farmactivities

Nonfarmactivities

Goal


However, household members have certain degrees of specialization and time alloca-tion. Members receiving the highest wage offers and employment opportunities willnaturally specialize in market work. Since men often receive better education andtraining, they develop more skills, receive higher wages, and have greater access toopportunities in the labor market than women. Thus, women tend to do more unpaidand home-based activities, which are often undercounted and undervalued. If theyare hired as seasonal agricultural laborers, they receive lower wages than men. Thus,in a particular farming system, even where the household is a unit of analysis, thepatterns of activities, resources, and participation in decision making are importantinformation and must be determined by investigation, not assumption (Feldstein et al1989).

The purpose of doing on-farm research is to generate more appropriate tech-nologies under farmers’ conditions, to raise the welfare of farm families, and to en-hance society’s goals. In conducting on-farm research, it is necessary to understandthe operation of the small farm within the wider context (village, district, regional,and global). Solving farmers’ problems requires an interdisciplinary team of scien-tists and active participation of the farmers, including women. The social scientistplays an important role at the beginning of the research process, in characterizing thesocioeconomic and cultural components of the livelihood systems, in identifying con-straints and opportunities that are consistent with the needs of the clients and users(men and women) of technologies, and in analyzing the ex ante and ex post impact oftechnologies.

Gender analysis in on-farm research

On-farm research with a systems perspective is an iterative, overlapping, and dy-namic process. However, any on-farm research should follow a process and can besummarized into these main activities:

● Characterizing/describing the local farming systems and practices and diag-nosing farmers’ problems and constraints to productivity,

● Selecting and improving existing technologies and techniques to overcomethese constraints, and

● Testing and adapting gender-responsive technologies with farmer participa-tion.

Using the same framework, gender analysis and women’s concerns can be inte-grated from the beginning of the research process and not only during the impactanalysis. The research can be divided into two phases:

● Identifying gender roles in the household, in on-farm and nonfarm activi-ties, and in farming practices; quantifying the labor contributions of menand women in major farm activities; assessing gender differences in accessto and control of resources; identifying constraints, potential, and needs; andidentifying options for improvement.

● Designing, testing, and adapting the best available technologies to increasewomen’s productivity and income and to reduce work burden.

344 Paris et al

Women and not men only should be included as cooperators in on-farm re-search. Their perceptions, needs, knowledge, and skills should be considered in thedesign, testing, and evaluation of proposed technologies. Opportunities or optionsthat can enhance women’s roles in terms of increasing productivity of labor and land,providing income opportunities, reducing their work burden, or enhancing their tech-nical knowledge or skills should be addressed in any research project on rice-basedfarming systems dealing with food security and poverty alleviation.

Application of gender analysis in rainfed lowland rice ecosystems:a case study in eastern Uttar Pradesh, India

Selection of research sites and respondentsThis case study was conducted in major rice-growing villages (Chandpur, Mungeshpur,and Sariyawan) in Faizabad District in eastern Uttar Pradesh. These villages are sitesof the Rainfed Lowland Rice Research Consortium coordinated by IRRI and the In-ternational Fund for Agricultural Development (IFAD)-funded IRRI-ICAR (IndianCouncil for Agricultural Research) Collaborative Project on Rainfed Rice in EasternIndia. This study is being conducted in collaboration with the Narendra Deva Univer-sity and Agricultural Technology (NDUAT) in Kumarganj, Faizabad District. Thesevillages differ in proximity to the major market (Faizabad City), agroecology, accessto supplementary irrigation, and amenities (Table 1). Chandpur is near Faizabad City,whereas Mungesphur and Sariyawan (adjacent villages) are far from it. Farmers haveheterogeneous land types—lowland, medium land, and uplands. More than half ofthe total cultivated area in Chandpur and Mungeshpur is lowland and medium land.In contrast, half of the total cultivated land in Sariyawan is under uplands. The popu-lation to land ratio is higher in Chandpur than in the other two villages. Farminghouseholds in Mungeshpur and Sariyawan have greater access to supplementary irri-gation for rice and nonrice crops such as wheat, potatoes, and vegetables than inChandpur.

A socioeconomic household survey was conducted in 1995-96 based on totalenumeration of the population. This survey covered 431 farming and landless house-holds. Different data collection methods were used starting with informal interviewsand a participatory rural appraisal (PRA) followed by household surveys and focusedgroup interviews. Within a household, the principal male or female was intervieweddepending upon the questions being pursued. However, for the labor data in rice pro-duction, the principal male and female household members who were actively in-volved in rice farming were interviewed together. The following section describes theenvironmental (social, cultural, agroclimatic, physical) characteristics for which gen-der roles and gender relations were analyzed.

The social and cultural environment that affects gender relationsAny attempt to assess the problems of Indian women in agriculture and their con-straints to improving the performance of farming systems is incomplete without alook at the social structure, cultural norms, and value systems that define and deter-


mine the roles of men and women (Mukhopadhyay 1984). Patriarchal ideology, dowryduring marriage, caste structure, and combined families are the social and culturalpractices that influence gender roles and gender relations, which in turn affect theintrahousehold resource allocation.

Patriarchal ideology. The predominant force in the social organization of In-dian society is patriarchy. Until 40 years ago, the legal status of Hindu women inIndia was based on laws dictated by the ancient Hindu lawgiver, Manu (first andsecond centuries AD). Briefly, the essence of Manu’s thesis was that women are sup-posed to continuously remain under some male authority, first under that of the father,then of the husband, and finally of the son. He prohibited widow remarriage, insti-tuted childhood marriage for girls, established the concept of dowry, and disinheritedwomen from their husband’s and father’s property. Though the Hindu Civil code of1956 has tried to eliminate many of these disabilities, the women’s movement inIndia still has a long way to go to rid the society of its oppressive customs (Ghosh1987).

Land ownership. In India, the acquisition, ownership, and transfer of propertyare through the male members of the family. Women have little access to ownershipof land or other productive assets because of Hindu Inheritance Law, which entailspatrilineal transmission of property. Although the right to inherit property inpostindependence India had been assured to female members by law, the law itself isa compromise with the traditional position, which does not recognize a female’s rightto ownership of property. Furthermore, the socialization of girls within the partrilinealform of social organization ensures that women will not be in a position to claim theirlegal rights. A woman’s lack of education, lack of legal knowledge, and dependence

Table 1. Village characteristics of the research sites, 1995.

Descriptors Chandpur Mungeshpur Sariyawan

Distance to Faizabad City (km) 3 km (near) 28 km (far) 28 km (far)Agroecology Shallow lowland and Shallow lowland Upland and

submergence-prone and drought-prone drought-proneTotal cultivated area (ha) 54 62 70Land types (%)

Upland 27 34 50Medium land 15 14 17Lowland 50 52 33

Total number of households 200 150 120Total population 1,244 802 700

Male 643 429 364Female 601 373 336Male/female ratio 107/100 115/100 108/100

Population/land (persons ha–1) 23 13 10No. of primary schools 1 1 1No. of preparatory schools 1 – 0No. of tube wells 7 15 12No. of shops 20 10 5

346 Paris et al

on the men in the family will prevent her from claiming them (Mukhopadhyay 1984).The bias against women in terms of land inheritance is also true in the Philippines andother countries that are patrilineal societies. Quisumbing’s (1995) study in the Philip-pines revealed that better-educated fathers also favor daughters in terms of education,whereas mothers with more land tend to favor sons. Without a title to land as collat-eral, women have also been excluded from institutional credit and are thus unable tosecure tools and capital for self-employment except through the more costly informalcredit system. The loss of land inheritance rights for widows leads to destitution anddependency on other people, especially the in-laws (Agarwal 1998).

Dowry during marriage. In India, giving material wealth (referred to as dowry)along with the bride in marriage is a customary practice among the higher caste Hin-dus, especially among the families with land. The bride’s family gives the bride clothesand jewelry and also presents the groom’s family with costly gifts, household goods,cash, and, in some instances, property. Greater economic wealth has resulted in anincrease in the amount of dowry among groups traditionally practicing it. It has alsoadversely affected women’s status through its institution in groups that formerly didnot subscribe to it. Traditionally, the scheduled castes did not practice dowry. Instead,the institution of bride-wealth was the custom, in which the groom’s family usuallygave the bride’s family some gifts and cash at the time of marriage. This tradition diedout many years ago when dowry became a status symbol and a reflection of economicwealth (Ghosh 1987). This practice negatively affects those with more female mem-bers as they are forced to give up their valuable resources in agriculture such as landand livestock. Households with daughters are forced to sell their livestock to raise thedowry requirements imposed by the parents of the groom during marriage. Thus,having daughters becomes a liability to a poor household as social norms dictate thatthey should get married at a certain age. The pressure of raising the dowry is borne bythe parents of the bride. Despite the existence of a law that prohibits the dowry sys-tem, this practice is deeply embedded, even in very poor families with limited assets.

Caste structure. The caste structure forms the basic foundation of India’s socialstructure. Caste is the official social stratification, which is defined since birth. Thelower castes are officially classified as the backward and scheduled castes. They areconsidered the most deprived and underprivileged in terms of access to resources andsocial status. To reduce the gap between the upper and lower caste, they are registeredin a special governmental schedule and are now entitled to certain statutory measuresof positive discrimination, such as reservation of seats in school and colleges, and jobreservation. Women among the lower castes are also given reserved seats in the localpolitical organization. Within the lower castes, the backward castes have a higherstatus than the scheduled castes on the social ladder. Caste is further subdivided intosubcaste groups. The predominant upper castes are the Pandeys and Singhs. Amongthe backward castes, subcastes are the Bhuj, Chauhan, Gupta, Kohar, Vishwakarma,Barber, and Yadavs. The scheduled castes include subcastes such as Harijans, Kori,and Pasi. The caste classification explains the occupation of the households in thevillage. For example, the Yadavs of the backward caste in Chandpur are known for


taking care of dairy cattle. On the other hand, the Harijans are the marginal smallholdersand landless workers whose main livelihood depends on agricultural labor.

People of the same caste tend to cluster themselves in the village. This is due tocaste relations based on the principle of impurity, a principle that permeated all as-pects of life, whether it was food, occupation, or other intercaste relations. The uppercastes were deemed more pure and the lower castes less pure or even impure. There-fore, the upper castes scrupulously observed marriage and commensal restrictionsand avoided physical contact with the lower castes. People could eat and drink withmembers of the same caste above them, but would not do so with the lower caste as itwas thought to be “polluting.” Similar restrictions affect the sharing of utensils. Seg-regation of castes is almost complete in matters of residence.

The extent of female participation in production in India is determined by anexus of class/caste hierarchy and norms of patriachal ideology. Women from theupper castes stay in seclusion or “indoors” and do not engage in manual work tomaintain their social status. Women from the lower castes have more freedom towork on their own farms and outside their homesteads to earn a living.

Combining families into one household. A household constitutes more than onefamily or a group of persons normally living together and taking food from a com-mon kitchen. The male head of the household is the principal male member. A typicalnuclear household consists of a husband and wife with their own children, whereas acombined family includes grandparents, the older children, and grandchildren. Strongkinship ties provide safety nets for farming households, especially during periods ofstress. The combined households still exist in India, particularly among the uppercastes that own large farm holdings. Families pool their resources, particularly labor,and jointly manage their farms. For example, one hectare of land may be subdividedfor the sons while all the earnings are given to the head of the household. One advan-tage of the combined family is that, when one family member is in need, all the othermembers help. However, because of the increasing population pressure and subdivi-sion of lands and family conflicts, combined households are breaking down into nuclearunits. The further subdivision of lands through inheritance has resulted in a small andmarginal size of landholdings, especially among the poor.

In India, it is customary for the bride to live with her in-laws, who manage acombined family consisting of sons, daughters, their in-laws, and their grandchildren.Wolf (1996), as cited in Kandiyoti (1991), argues that the key to the reproduction ofclassic patriarchy lies in the operations of the patrilocally extended households, whichare commonly associated with the reproduction of the peasantry in agrarian societies.Under classic patriarchy, girls are given away in marriage at a very young age intohouseholds headed by the husband’s father. There, they are subordinate not only to allthe men but also to the more senior women, especially their mother-in-law.

Characteristics of the householdsFarming is the major source of livelihood in the two remote villages. The proportionof farming households is highest in the remote villages Sariyawan (100%) andMungeshpur (89%) and lower in the nearer village, Chandpur (76%). The majority of

348 Paris et al

households belong to the lower caste having marginal and small landholdings. Com-bined or extended households are still prevalent in Chandpur and Sariyawan (Table2).

A majority of the farm households are headed by males. However, the propor-tion of de facto female heads of households is higher in Sariyawan (27%) than inChandpur (19%) and Mungeshpur (11%) (Table 2). A wife may act as the de factohead of the household when she makes most of the decisions in the household and onthe farm. This happens when the husband is an invalid or sick or when he worksoutside the village for seasonal or permanent employment. It is a common practice inboth villages for husbands to leave the village to find seasonal employment duringthe slack periods on the farm (i.e., usually in December and January after sowing ofwheat) and come back in June during land preparation of rice. The male farm opera-tors in Chandpur and Mungeshpur are generally in their forties, but not in Sariyawan,where they are in their fifties. Wives are younger than their husbands. The averageeducational level of the wife is lower (1 to 2 years) than that of her husband (4 to 7years).

Table 2. Socio-demographic indicators by village (percentage andmean), 1995.

Indicators Chandpur Mungeshpur Sariwayan(near) (far) (far)

Total number in household 200 150 81Type of household (%)

Farming/landed 76 89 100Landless 25 11 –

Caste (% of households)Upper 10 8 7Backward 48 50 53Scheduled 42 42 40

Kinship (% of households)Nuclear 47 61 30Extended 54 39 70

Household head (% of households)Male-headed 75 81 69De facto female-headed 19 11 27De jure female-headed 7 8 4

Average household size by casteUpper 8 7 6Lower 9 6 7

Average age (years)Male operator 45 42 52Wife 41 37 48

Average years in schoolMale operator 7 7 4Wife 2 1 1


The biophysical environmentThe performance of the farming system in these typical rainfed villages is affected byunpredictable and erratic rainfall resulting in drought, waterlogging, or submergence;nutrient deficiencies and toxicities; and weeds, insects, and disease pressure (Singh1996).

Erratic rainfall pattern. The problem with the rainfall pattern is that it is erratic,sometimes too much and sometimes none, and hence the floods and droughts thatoccur are major sources of crop loss (Widawsky and O’Toole 1990). Rice suffersfrom stress during the vegetative stage because of too little water in June or too muchwater during the panicle initiation stage in August. These uncertainties pose particu-lar hardships for the poor, who face chronic vulnerability in terms of access to re-sources (Hossain 1995). In fact, the lives of the poor in India have been characterizedby the almost total absence of security (Dreze and Sen 1989). The years 1995 and1996, when this study was conducted, were drought years in Faizabad District, withdrought being severe in 1995 (Pandey et al 1997).

Heterogeneous land types and moisture stress. Aside from the unpredictablerainfall that makes farming very risky, the rainfed lowland farms suffer from differenttypes of stress. Results using geographic information systems showed that, of thetotal rainfed area of Faizabad, 35% is drought-prone, 40% is shallow favorable, whereasthe rest has problems of drought or submergence, or both, as well as sodicity. Chandpurrepresents a shallow and submergence-prone area, which is favorable rainfed duringthe years of low rainfall, whereas Mungeshpur and Sariyawan represent a drought-prone area that is favorable rainfed during the years of high rainfall (Singh 1992).Moreover, farmers have heterogeneous land types (lowland, medium, and upland)that affect rice varietal choice and performance.

Thus, improved crop establishment methods and the introduction of short- andmedium-duration rice varieties that depend on the land types are some of the tech-nologies that have to be introduced to these villages to minimize the risks of growingrice during the kharif season. Within a farming systems framework, rice improve-ment must focus on increased tolerance for the predominant abiotic and biotic stressessuch as late-season drought, submergence for less than 10 days, and blast (Sarkarung1996).

Cropping patternsDepending on rainfall distribution, the crop year is divided into three growing sea-sons: the kharif or monsoon season (June-July to November-December), the rabi orwinter season (November-December to March-April), and the zaid or summer season(March-June). In the lowlands and medium lands, rice-wheat and rice-wheat mixedwith mustard were the predominant cropping patterns. Rice is planted in June-Julyand harvested in October or November, depending on the growth duration of thevarieties used. To take advantage of residual moisture from the soil, farmers broad-cast wheat immediately after harvesting rice in November. Most of the farmers broad-cast mustard seeds after sowing wheat. In wheat + mustard intercropping, mustard isusually harvested early, in the first or second week of March. Wheat is harvested from

350 Paris et al

the last week of March until the second week of April. If rice is harvested late, wheatis also sown late and, consequently, wheat is not mixed with mustard. A few farmersgrow wheat and mustard in separate fields because they think that mixing these cropslowers wheat yields. Two types of mustard seeds are grown. The first type is Lahi orToria (60–80-day variety), which is grown after rice or maize in the uplands or afterCurbi (green fodder crop). This variety is planted in October and harvested in Decem-ber as a sole crop. The second type is Varuna (120-day yellow variety), sown withwheat, gram pea, berseem, and potato. Mustard provides household oil needs asidefrom being a high-value feed (mustard oil cake) for livestock. It is also an importantsource of cash and income. For thinning purposes, farmers remove mustard as greenfodder for animals and also as a green leafy vegetable earlier in the season. Dry mus-tard straw is also used for roofing and fuel (Fig. 2).

In the medium lands and lowlands, pea, gram, and lentil are planted in Octoberafter rice and harvested in March. Farmers who have access to supplementary irriga-tion and raise livestock also grow berseem, a fodder crop, during this period. In theuplands, farmers grow sugarcane and pigeonpea throughout the year. Thus, farmersmaximize the use of small plots by combining wheat and mustard to meet their food,fuel, oil, and animal feed requirements and for other housing materials.

Seasonal calendar and gender division of laborThe seasonal calendar provides a format for analyzing activities by season and bygender. This seasonal calendar identifies the busy and slack months and the patternsof activities by male and female labor. It identifies “who does what,” particularly asthis relates to the agricultural year and other seasonal patterns (Table 3). There aregender-specific tasks or degrees of specialization in rice production. Males are exclu-sively responsible for preparing the land, broadcasting seeds, and applying chemical

Fig. 2. Rainfall and cropping pattern in Chandpur and Mungeshpur.

350

300

250

200

150

100

50

0

Rice

Wheat/wheat + mustard

Pea/gram/lentil/berseem

Medium +lowland

Upland andmediumirrigatedPigeonpea

Sweet potato

Potato

Lahi (oilseed)Vegetable

Sudan chari/maize

Sugarcane

Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May

Rainfall (mm)

1995

1996

Curbi (local fodder)

Rice


Tabl

e 3

. S

easo

nal c

alen

dar

and

gend

er d

ivis

ion

of la

bor,

1995.

Maj

or a

ctiv

ities

June

July

Aug

Sep

tO

ctN

ovD

ecJa

nFe

bM

arAp

rilM

ay

Kha

rif (

July

/Jul

y-N

ov/D

ec)

Prep

arin

g la

nd f

or n

urse

ry s

eedb

edM

MAp

plyi

ng f

arm

yard

man

ure

on fie

lds

F>M

aF

Irrig

atin

g ric

e fie

lds

MPu

lling

see

dlin

gs a

nd t

rans

plan

ting

rice

FF

Wee

ding

ric

e fie

lds

FF

Har

vest

ing

rice

F>M

F>M

Thre

shin

g ric

eF>

MF

Han

d-po

undi

ng r

ice

F

Rab

i (N

ov/D

ec-M

ar/A

pril)

Land

pre

para

tion

for

pota

to,

oils

eeds

, pe

aM

MS

owin

g po

tato

, oi

lsee

ds, pe

a, a

nd b

erse

em (fo

dder

)M

>F

Land

pre

para

tion

for

whe

atM

>F

Sow

ing

whe

atM

M>

FIr

rigat

ing

and

appl

ying

fer

tiliz

er o

n po

tato

fie

lds,

harv

estin

g oi

lsee

dsM

>F

Har

vest

ing

pota

to a

nd m

usta

rdF

FH

arve

stin

g pe

a an

d m

usta

rdF

Har

vest

ing

whe

at, pi

geon

pea,

gra

mF

FTh

resh

ing

rabi

cro

psM

>F

Oth

er a

ctiv

ities

Mak

ing

cow

dun

g ca

kes

FF

FF

FF

Non

farm

wor

kM

MM

MM

Col

lect

ing

anim

al fod

der

FF

FF

FF

FM

akin

g ba

sket

s/st

orag

e bi

nsF

FAn

imal

car

e (m

ilkin

g, f

eedi

ng)

F>M

F>M

F>M

F>M

F>M

F>M

F>M

F>M

F>M

F>M

F>M

F>M

a F>M

mea

ns t

hat

fem

ale

part

icip

atio

n is

mor

e th

an m

ale

part

icip

atio

n.

352 Paris et al

fertilizers, whereas women are predominantly responsible for transplanting, weed-ing, harvesting, hand-threshing, hand-pounding, and cooking rice. Starting in June,male farmers start preparing the land and the nursery seedbed for rice production.July is the busiest month for women, when they are involved in applying farmyardmanure on the rice fields, pulling rice seedlings from the seedbeds, and transplantingthem on puddled fields. In August, women spend most of their time weeding ricefields. They continue these tasks in September. In October and November, womenharvest rice and thresh until December. During the rabi season from September toOctober, men prepare the land for potato, Lahi, and pea depending on the time ofharvesting of rice. During this period, women start to make dry cow dung cakes forfuel. In December, after sowing wheat, men begin to leave the villages in search ofwork. Women are busy weeding plots planted to vegetables and spices, collectingfodder for animals, hand-pounding rice, making storage bins out of clay, making cowdung cakes, and making baskets from local materials. In January, men irrigate thepotato fields and tend the other rabi crops before they leave for nonfarm jobs. Womencollect animal fodder, harvest mustard, and continue to make cow dung cakes forfuel. Women harvest potato and mustard in February and continue to make cow dungcakes and collect animal fodder. In March, women are engaged in harvesting potato,gram, and mustard from the wheat fields. May is the only month wherein women arenot involved in field activities.

Rice productionRice is the major crop grown in the three villages. It occupies 60%, 73%, and 57% ofthe total cultivated land in Chandpur, Mungeshpur, and Sariyawan, respectively. Othercrops such as sugarcane, pulses, pigeonpea, vegetables, and fodder crops are alsogrown on upland fields during the kharif season. About 10% of the total land is leftfallow. There is more diversity in crops grown after rice during the rabi season (Table4). In these villages, the adoption of improved varieties is high, ranging from 82% to93%; however, despite their high adoption, average yields are low, ranging from 1.9to 2.4 t ha–1, and only slightly higher than that of the local varieties, which averageless than 1.5 t ha–1 (Table 5). Pandey et al (1998) in a similar study in eastern UttarPradesh revealed that yield is variable from plot to plot due to differences in soil typeand management practices. Farmers, especially in Mungeshpur, continue to grow tra-ditional varieties because of their tolerance for submergence and drought comparedwith the improved varieties. Sarkarung (1996) mentioned that the majority of im-proved rice cultivars developed on-station failed to perform under farmers’ condi-tions. This implies that the newly released cultivars could not compete with nativeand traditional cultivars under adverse conditions where water and fertility are notcontrolled. Aside from the heterogeneity in land types and management, farmers havedifferent uses, needs, and preferences for rice varieties based on their socioeconomicdifferences, which affect varietal adoption. Thus, farmers’ criteria for rice varietalselection in the rainfed rice environments have to be well understood by both plantbreeders and social scientists.


Table 4. Crops grown during the kharif and rabi seasons, 1995.

Chandpur Mungeshpur Sariyawan

Crops Total % of Total % of Total % ofarea total area total area total(ha) area (ha) area (ha) area

KharifRice 24 60 40 73 32 56Pigeonpea 3 8 3 5 – –Vegetables 1 3 2 4 3 5Curbi 4 10 3 5 – –Sugarcane 4 10 2 4 3 5Pulses – – – – 8 14Pulses + vegetables/others – – – – 2 3Others – – – – 5 9Fallow 4 10 5 9 4 7Total 40 100 55 100 57 100

RabiWheat + mustard 27 44 40 62 25 44Wheat + pulses – – – – 8 14Wheat 7 11 2 3 1 2Pulses – – – 10 18Pigeonpea 5 8 8 12 – –Mustard (oilseed) 3 5 – – 1 2Green fodder (berseem) 2 3 – – – –Vegetables 4 7 3 5 2 4Spices – – – – 1 1Sugarcane 4 7 6 9 3 6Pea + mustard 5 8 4 6 – –Oilseed + vegetables/spices/others – – – – 1 1Vegetables + spices/others – – – – 4 6Others – – – – 1 2Fallow 4 7 2 3 – –Total 61 100 65 100 57 100

Table 5. Rice yield and adoption of rice variet-ies, 1995.

Actual Normal % of totalVariety yield yield rice area

(t ha–1) (t ha–1)

ChandpurAll improved 2.0 2.5 83All local 1.5 2.1 17

MungeshpurAll improved 2.4 3.0 82All local 1.1 2.3 18

SariyawanAll improved 1.9 2.5 91All local 1.3 1.9 9

354 Paris et al

Labor use in rice production and gender division of laborRice production is a very labor intensive activity that employs both men and womenin production until postharvest activities. Average labor days per hectare on rainfedlowland environments from key sites in South and Southeast Asia range from 95 to270 d depending on the variety used, levels of technology, management, etc. (Table6). Female labor participation in rice production can vary by country, agroecosystem,class/caste, availability of male labor, mother’s stage in the life cycle, and other fac-tors. Compared with that of males, female labor participation is highest in Lao PDRand India, more than half in Thailand, Vietnam, and Nepal, and less than half in Indo-nesia and the Philippines. A closer look into labor days per hectare in four villages inFaizabad District reveals a higher proportion of female participation in the villagesnear the cities (Chandpur and Khanpur) than in the remote villages (Mungeshpur andSariyawan) where farming is the major source of livelihood.

Women are not a homogeneous group; rather, they belong to different socio-economic categories such as class and caste. Caste, which is positively correlatedwith farm size, determines the extent of female labor participation in rice productionin eastern Uttar Pradesh. A focus interview from households sampled in the socioeco-nomic surveys was conducted to quantify the labor inputs in rice production of adultmales and females from different sources, by operation and by social status. Amongthe upper caste, female family members do not provide labor on their own farms, butinstead hire women from the lower caste to substitute for their labor. However, withinthe confines of their homesteads they select and store seeds and feed the animals. Theupper castes follow a very strict system of seclusion (purdah) (Table 7). In Chandpur,

Table 6. Labor inputs in rainfed rice production (d ha–1), 1995.

Country Villages Total Male Female(d ha–1) (%) (%)

Indonesia Jakenan, Central Java 161 54 46Sumber, Central Java 178 59 41

Thailand Ban Sai Khram, South Thailand 104 45 55Ban Don Paw Daeng 102 46 54

Philippines Carosucan, Sta. Barbara 133 83 17Tampac, Nueva Ecija 188 68 32

Cambodia Kandal and Takeo 167 54 46Vietnam He Thu District 105 45 55Lao PDR Khok Nghai, Xaythani District 110 24 76

Ak-sang, Phonethong District 117 38 62Nepal Naldung, Nagarkot (midhills) 269 42 58

Mohana, Rantnagar (lowland) 101 50 50Baghmara, Rantnagar (lowland) 95 45 55

India Chandpur, Faizabad District (near) 187 16 84Khanpur, Faizabad District (near) 210 24 76Mungeshpur, Faizabad District (far) 132 33 67Sariyawan, Faizabad District (far) 211 45 55

Sources: IRRI (1990, 1992).


Table 7. Labor input (person-days ha–1) in rice production, Chandpur, Faizabad District, easternUttar Pradesh. Use of exchange labor is not practiced in this village, 1995.

Family Hired TotalCaste/operation Total

M F M F M F labor

Upper caste (n = 11)Prepare land 1.0 0.0 2.2 0.0 3.1 0.0 3.1Pull seedlings/transplant 0.0 0.0 5.8 62.7 5.8 62.7 68.5Broadcast seeds 2.0 0.0 0.7 0.0 2.8 0.0 2.8Apply fertilizer 2.4 0.0 1.7 0.3 4.1 0.3 4.3Weed 0.0 0.0 0.0 38.9 0.0 38.9 38.9Irrigate 2.1 0.0 0.5 0.0 2.6 0.0 2.6Harvest 1.4 0.0 2.3 26.6 3.6 26.6 30.2Thresh (manual) 0.0 0.3 0.5 29.4 0.5 29.7 30.3

Total 8.9 0.3 13.7 157.9 22.5 158.2 180.7Percentage of total labor 4.9 0.2 7.6 87.4 12.5 87.5 100.0

Backward caste (n = 51)Prepare land 6.3 0.0 0.4 0.0 6.8 0.0 6.8Pull seedlings/transplant 4.2 14.3 0.7 39.0 4.9 53.3 58.1Broadcast seeds 0.9 0.0 0.2 0.0 1.1 0.0 1.1Apply fertilizer 6.8 3.6 0.0 0.0 6.8 3.6 10.5Weed 2.7 24.1 0.8 30.3 3.4 54.4 57.8Irrigate 0.4 0.1 0.0 0.0 0.4 0.1 0.5Harvest 9.7 17.9 0.1 7.1 9.8 25.0 34.7Thresh (manual) 0.6 29.8 0.0 7.3 0.6 37.1 37.7


Scheduled caste (n = 31)Prepare land 4.4 0.0 2.2 0.0 6.5 0.0 6.5Pull seedlings/transplant 3.5 19.9 0.8 34.3 4.3 54.2 58.5Broadcast seeds 0.3 0.0 0.7 0.2 1.0 0.2 1.1Apply fertilizer 8.2 6.4 0.1 0.1 8.3 6.5 14.8Weed 1.9 36.0 0.0 21.5 1.9 57.5 59.4Irrigate 6.5 2.2 0.1 0.0 6.6 2.2 8.7Harvest 7.4 24.8 1.0 5.0 8.4 29.8 38.2Thresh (manual) 2.4 35.0 0.7 4.3 3.0 39.3 42.3


356 Paris et al

female family members among the upper caste households contribute less than 1% oflabor. In contrast, female members among the backward and scheduled castes con-tribute 43% and 54%, respectively. These results are similar to the study of Kelkar(1992) on the roles of women in Bihar, which are influenced by class and caste fac-tors. In this village, the contributions of upper and lower caste male family membersto total labor in rice production are 5% and 15%, respectively. The low participationof male family members in farming is due to their participation in nonfarm employ-ment in Faizabad City. Most male members commute every day to the city for workand leave most of the farm work to the female family members.

Similarly, in Mungeshpur, female family members from the upper castes do notwork on their own farms. In contrast, the female family members from the lowercastes contribute about one half of the total labor input in rice production. Among theupper caste, the male family members contribute a low 7%, whereas the men from thelower caste contribute about one-fourth (Table 8). Although upper caste women arenot supposed to provide physical labor in crop production activities outside the home-steads, our findings show that women from the upper caste in Sariyawan village breakthe norms, out of economic necessity. Female family members from the upper castecontribute about 10% of labor in rice production (Table 9). These women are eitherwidows or de facto heads of households who are left to manage their own farms.

Exchange labor is still commonly practiced in Mungeshpur and Sariyawan, as astrategy for managing production requirements when cash is scarce. About 20 house-holds from the lower caste organize themselves so that they can exchange labor. Asfound by Chen (1990), households from the same kinship and caste group often bor-row, loan, pool, or exchange productive assets, including labor, pump sets, farm imple-ments, and bullocks.

Value of unpaid male and female labor in rice productionWomen’s unpaid work in rice production can be made “visible” by imputing the valueof male and female family labor using the prevailing market wage rates. An estima-tion of noncash costs in rice production (Table 10) reveals that female family labor(exchange and hired) contributes about 25% of the total costs. In contrast, male fam-ily members contribute less than 10%. On a per-hectare basis, the imputed value ofthe unpaid labor of female family members is about US$35.71 per hectare. Theseresults indicate the importance of female family labor in saving the costs of hiringlabor in rice production. However, the work burden of women from the lower caste,small, and marginal farming households increases without necessarily having a com-pensatory improvement in their standard of living. Beteille (1985, cited in Ghosh1987) has observed that most Western scholars associate emancipation of womenwith the right to work. But there is also another side to the picture. In agrarian societ-ies, work is regarded more often as a hardship than as a privilege, because agricul-tural work entails arduous physical work, low status, and low pay or no pay as in thecase of female family members (Ghosh 1987).


Table 8. Labor input (person-days ha–1) in rice production, Mungeshpur, Faizabad District, east-ern Uttar Pradesh, India, 1995.

Family Exchange Hired TotalCaste/operation Total

M F M F M F M F labor

Upper caste (n = 5)Prepare land 0.0 0.0 0.0 0.0 3.4 0.0 3.4 0.0 3.4Pull seedlings/transplant 0.3 0.0 0.0 0.0 17.3 32.4 17.6 32.4 50.0Broadcast seeds 0.5 0.0 0.0 0.0 0.3 0.1 0.8 0.1 0.9Apply fertilizer 4.1 0.0 0.0 0.0 0.8 0.0 5.0 0.0 5.0Weed 0.1 0.0 0.0 0.0 8.1 27.6 8.3 27.6 35.8Irrigate 2.6 0.0 0.0 0.0 0.0 0.0 2.6 0.0 2.6Harvest 2.2 0.0 0.0 0.0 7.6 20.2 9.8 20.2 30.0Thresh (manual) 1.0 0.0 0.0 0.0 4.7 16.3 5.6 16.3 21.9

Total 10.8 0.0 0.0 0.0 42.2 96.6 53.1 96.6 149.6Percentage of total labor 7.2 0.0 0.0 0.0 28.3 64.5 35.5 64.5 100.0

Backward caste (n = 33)Prepare land 9.6 0.0 0.0 0.0 0.6 0.0 10.1 0.0 10.1Pull seedlings/transplant 6.3 16.2 1.6 4.3 6.2 15.7 14.0 36.1 50.1Broadcast seeds 0.4 0.1 0.0 0.0 0.0 0.0 0.4 0.1 0.5Apply fertilizer 3.5 2.7 0.0 0.0 0.1 0.0 3.6 2.7 6.2Weed 3.0 14.4 0.2 2.6 1.4 5.1 4.6 22.1 26.7Irrigate 3.1 1.0 0.0 0.0 0.2 0.0 3.3 1.0 4.3Harvest 5.5 15.6 0.5 3.3 1.2 3.8 7.3 22.7 29.9Thresh (manual) 3.2 20.9 0.0 2.3 0.8 2.3 4.0 25.6 29.5


Scheduled caste (n = 28)Prepare land 8.4 0.0 0.0 0.0 0.8 0.0 9.2 0.0 9.2Pull seedlings/transplant 8.1 20.1 0.8 3.6 4.1 14.2 13.0 37.9 50.9Broadcast 0.4 0.1 0.0 0.0 0.0 0.0 0.4 0.1 0.4Apply fertilizer 4.4 5.3 0.0 0.0 1.0 0.0 5.3 5.3 10.6Weed 7.7 22.9 0.0 1.3 1.0 6.3 8.7 30.5 39.2Irrigate 3.9 2.5 0.0 0.0 0.0 0.1 3.9 2.6 6.5Harvest 8.3 18.5 0.0 0.8 0.9 4.1 9.2 23.4 32.6Thresh (manual) 5.6 22.2 0.0 0.8 0.1 0.9 5.7 23.9 29.6


358 Paris et al

Table 9. Labor input (person-days ha–1) in rice production, Sariyawan, Faizabad District, easternUttar Pradesh, India, 1995.

Family Exchange Hired TotalCaste/operation Total

M F M F M F M F labor

Upper caste (n = 4)Prepare land 6.4 0.3 0.0 0.0 5.3 0.0 11.6 0.3 11.9Pull seedlings/transplant 2.5 1.3 0.0 0.0 13.8 47.8 16.3 49.0 65.3Apply fertilizer 3.6 0.0 0.0 0.0 0.0 0.0 3.6 0.0 3.6Apply FYM 7.6 0.6 0.0 0.0 4.6 0.0 12.3 0.6 12.9Weed 6.5 5.0 0.0 0.0 1.0 13.4 7.5 18.4 25.9Irrigate 5.1 0.9 0.0 0.0 0.0 0.0 5.1 0.9 6.1Harvest 6.6 5.6 0.0 0.0 5.6 24.4 12.3 30.0 42.3Thresh 5.4 4.7 0.0 0.0 4.3 13.9 9.6 18.6 28.2


Backward caste (n = 9)Prepare land 15.7 0.8 0.0 0.0 1.2 0.0 16.9 0.8 17.7Pull seedlings/transplant 14.7 17.2 0.7 4.2 2.5 10.4 18.0 31.8 49.8Apply fertilizer 5.1 0.0 0.0 0.0 0.0 0.0 5.1 0.0 5.1Apply FYM 5.7 3.9 0.0 0.0 1.0 0.6 6.7 4.5 11.2Weed 16.9 19.0 0.0 4.0 2.5 13.6 19.3 36.5 55.9Irrigate 1.3 0.1 0.0 0.0 0.0 0.0 1.3 0.1 1.4Harvest 12.6 18.2 0.0 1.2 1.0 3.1 13.6 22.6 36.2Thresh (manual) 7.5 15.5 0.1 0.3 0.2 2.5 7.8 18.2 26.0


Scheduled caste (n = 27)Prepare land 20.1 0.4 0.0 0.0 0.0 0.0 20.1 0.4 20.5Pull seedlings/transplant 17.8 17.9 0.0 5.4 0.0 0.0 17.8 23.3 41.1Apply fertilizer 5.4 0.0 0.0 0.0 0.0 0.0 5.4 0.0 5.4Apply FYM 4.7 2.9 0.0 0.0 0.0 0.0 4.7 2.9 7.6Weed 23.6 27.5 0.0 5.6 0.0 2.7 23.6 35.7 59.4Harvest 13.6 19.6 0.0 1.7 0.0 0.0 13.6 21.2 34.9Thresh 6.3 14.5 0.0 0.0 0.0 0.0 6.3 14.5 20.8



Animal systems

Across all the study villages, animals constitute an integral part of the mixed farmingsystems. Farming households raise a mix of a small number of bullocks, cattle, andbuffalo. Bullocks and male buffaloes are used as draft power for plowing and harrow-ing the fields, and transporting farm products. Animal manure is used as organic fer-tilizer for crops and converted into dung cakes for household fuel. On the other hand,the biomass and by-products of crops and residues from the fields are fed to the live-stock. Crop biomass includes straw of rice and wheat, green sugarcane tops, andpigeonpea and gram straw. Rice and wheat straw are also used as bedding, particu-larly during winter, as roof thatch, and as storage insulators, and are mixed with clayfor making storage bins. Rice straw usually lasts for 3 months. It is also a source ofcash income when sold during times of fodder scarcity. Farmers who grow sugarcaneuse bullocks to crush the sugarcane stalks. Female buffaloes are raised for milk, curd,and ghee, which are parts of the daily diet of Indian families.

Table 10. Cost and returns of rice production, Faizabad District, eastern UttarPradesh, 1995.


Variables US$ % US$ % US$ %

Cash inputsSeeds 7.57 5 7.37 5 6.89 5Fertilizer 31.94 19 26.43 18 28.86 20Tractor rent 22.17 13 8.57 6 6.49 5Animal rent 1.20 1 1.14 – 2.26 2Irrigation 30.06 18 28.91 20 17.71 12Hired male labor 0.29 – 2.26 2 2.51 2Hired female labor 18.00 11 4.66 3 7.14 5Total paid-out cost (TPC) 111.23 67 79.34 54 71.86 51

Noncash inputsFarmyard manure 6.60 4 2.60 2 2.91 8Family adult male 8.17 5 12.06 8 11.29 8Exchange adult male – – 1.71 1 2.00 1Family adult female 30.46 19 27.00 18 26.11 18Exchange adult female 3.23 2 8.63 6 6.54 5Own animal 3.97 2 11.09 8 10.31 7Exchange animal 0.71 1 3.71 3 2.89 2Total noncash cost (TNC) 53.14 33 66.80 46 62.05 49Total cost (TC) 164.37 100 146.14 100 133.91 100Gross returns (GR)a 274.57 214.26 238.97Net income (GR – TPC) 163.34 134.92 167.11Surplus (GR – TC) 110.20 68.12 105.06Average yield (t ha–1) 2.2 1.7 1.6Average rice area (ha) 0.23 0.29 0.34

aIncludes byproducts.

360 Paris et al

Ownership of livestock allows farmers to fend off risks of drought and to main-tain capital in the form of animals as insurance against bad times, particularly duringdrought. Livestock are also raised to cover large expenditures such as medical ex-penses and payment for dowry during marriage. Farmers also use livestock in ex-change for another resource. For example, a farmer with a tube well can providewater to the land of another farmer who, in turn, prepares his land with his pair ofbullocks. The landless laborers who own livestock but do not have crops and cropresidues work on a sharing basis and provide inputs to the land they cultivate as longas they get rice and wheat straw for their animals. For poor women, raising goatsprovides them with independent income, security, and instant cash during times ofemergency.

Gupta (1991) stressed that, in a rainfed economy, it is not the crops but thelivestock that are the main anchor of household survival in the dry regions. Once thisis recognized, the primacy of fodder (whether from grasslands, trees, or crop resi-dues) vis-à-vis grain becomes clear.

Animal populationA higher proportion of the farming households in Chandpur raises cattle and buffa-loes, whereas farming households in Mungeshpur and Sariyawan raise more bul-locks. Women from poor and landless households raising goats is a more popularactivity in Mungeshpur and Sariyawan. A few households raise pigs as a means ofsecurity and instant cash (Table 11). Because of the proximity of Chandpur to themarket, household members, particularly those belonging to the backward castes(Yadavs), raise dairy cattle for consumption and for sale. According to key informantinterviews, the bullock population in Chandpur has been declining because farmersare shifting to the use of tractors for land preparation.

Table 11. Percentage of farming households that own animals byvillage, 1995.

Chandpur Mungeshpur Sariyawan(n = 151) (n = 133) (n = 81)

AnimalNo. of No. of No. of

households % households % households %

Cattle 89 59 60 45 16 20Buffalo 101 67 42 32 36 44Bullock 13 9 38 29 44 54Goat 12 8 31 23 22 27Pig – – 3 2 2 2Chicken – – 2 2 – –


Gender roles in animal husbandryWhen one visits the villages in eastern Uttar Pradesh, it is common to see womencarrying a huge headload of green fodder home to where the animals (buffaloes orcows) are kept tethered in the courtyard. Since women are responsible for milkingand taking care of milch animals, they bear the responsibility for collecting and cut-ting green fodder for the animals. Twice a day, they feed and milk the animals, andalso clean the shed. With limited grazing lands, women on average spend half a daywalking long distances to collect grasses and weeds, particularly in Chandpur, wheremore dairy cattle require green fodder every day. Thus, women are the worst affectedwhen drought occurs because this means that they have to walk farther and spendlonger hours to collect animal fodder for their livestock (Paris et al 1998). Poor womenalso volunteer to weed the fields without wages as long as they can take home theweeds for their animals. During the summer season (November to February), womenspend 4 to 6 hours per day making cow dung cakes for household fuel. Making drydung cakes is an income- or expenditure-saving activity wherein women save ap-proximately US$20 per year. In Chandpur, five to six households sell cow dung cakesfor fuel. Because of poverty, women minimize the use of purchased inputs whilemaximizing the exploitation of residues, by-products, and their own labor. A declinein the animal population will result in increasing demands on women’s schedules, asthey will have to travel farther into the forests in search of fuel. On the other hand,when more animals are raised, women’s work burden is greater. Thus, technologiesthat can increase the availability of fuel and animal fodder will directly benefit thefemale members of farming households.

Diversifying sources of income

Despite the importance of rice as a staple food and in terms of land area, rice contrib-uted only a small (less than 12%) proportion of the total income in 1995-96 (Table12). Thus, farming households diversify their income sources. Wheat and other crops(pulses, oilseeds, sugarcane, vegetables, etc.) are the major crops in terms of theirshare in income, particularly in the remote villages. Farmers, particularly the lowercastes with small landholdings, rely more on nonfarm income. Sales of livestockproducts such as milk are also an important source of cash income among the farminghouseholds. For women, taking care of goats is one strategy for securing an indepen-dent source of income. They use their earnings to buy their own saris (clothes), bedsheets, food, and medicine for their families.

Family members from the lower caste, particularly the female members, derivecash by working as agricultural laborers on other farms. Thus, the higher the croppingintensity within the village, the greater is the employment and income opportunity forthe poor and landless women. Because of the lower wages received by women, how-ever, the proportion of off-farm income to total income is quite low.

362 Paris et al

Time allocation

Men and women have different uses of time depending on their degrees of specializa-tion and opportunity costs of their labor. To better understand the differences in timeuse, 20 principal males and females from lower caste farming households inMungeshpur were interviewed with regard to the number of hours they spend per dayand per month in major activities. The relative contribution of the principal females isconsistently higher than that of the principal males in all activities, except in nonfarmemployment (Fig. 3). The principal females spend 62% in farm activities, 61% in off-

Fig. 3. Time allocation of principal male and female family mem-bers, Mungeshpur village, Faizabad.

Table 12. Percentage share of different sources of income of farming households,by village and caste, 1995-96.

Chandpur Mungeshpur SariyawanIncome source

Upper Lower Upper Lower Upper Lower(n = 13) (n = 97) (n = 12) (n = 121) (n = 6) (n = 75)

Rice 7 5 9 11 9 4Wheat 5 4 11 13 8 9Other crops 7 5 21 9 42 46Livestock 35 13 14 15 13 7Farm by-products 2 1 10 2 – 4Rent of machine – 1 2 – 20 8Farm labor – 3 – 5 – 11Nonfarm 44 68 33 45 8 11

Total 100 100 100 100 100 100Av (US$ y–1) 2,090 1,100 1,260 600 860 200

Household

Animalmanagement

Nonfarmactivities

Off-farmactivities

Farmactivities

0 20 40 60 80

Percentage of total time used

Males

Females


farm activities, 25% in nonfarm activities, 75% in animal management, and 76% inhousehold activities. This indicates that women’s work burden is greater than that oftheir male counterparts in almost all activities except nonfarm work. Most often womenhave to combine reproductive and productive activities to meet the competing de-mands for their time.

Resource endowments

Farm size, rice area, land use, and rice diversity by social statusAlthough India has a rich agriculture, with a huge mass of land, farming is generallydominated by small landholders from the lower castes, whereas the large landholdersbelong to the upper castes (Gopalan 1992). This is evident in these study villageswhere farm size and rice area are negatively related to social status (Table 13). Thesmall size and fragmentation of landholdings are major constraints to increasing theefficiency of rice productivity, particularly in the allocation of water, land preparationwith the use of tractors, and proper management of plots. Farming households fromthe lower castes, however, tend to maximize their land use throughout the year asreflected by the cropping intensity indices (CII). Moreover, the lower castes tend to

Table 13. Indicators of land use and rice diversity, 1995.

Average RiceVillage/caste farm size area CIIa CDI RVI

(ha) (ha)

ChandpurUpper 1.01 0.49 161 0.76 0.40Backward 0.33 0.22 197 0.60 0.18Scheduled 0.24 0.17 198 0.60 0.13

MungeshpurUpper 1.32 0.56 175 0.77 0.44Backward 0.60 0.38 186 0.68 0.27Scheduled 0.26 0.22 196 0.60 0.13

SariyawanUpper 3.56 1.01 150 0.73 0.44Backward 1.56 0.40 150 0.66 0.38Scheduled 0.93 0.33 150 0.64 0.44

aCII (crop intensity index) indicates the extent of land use. An index of 200indicates full use of land. CDI (crop diversification index) indicates the diversityof crops grown. This index ranges between zero and one with higher valuesindicating a greater degree of diversification. RVI (rice diversity index) indicatesthe number of rice varieties grown. This index ranges between zero and one withhigher values indicating a greater degree of diversification.

364 Paris et al

meet their food and fodder requirements from their own limited land. Farming house-holds from the upper castes that have a larger size of landholdings tend to grow morethan one crop and more than one variety of rice as shown by the crop diversification(CDI) and rice varietal (RVI) indices.

Access to supplementary irrigationA majority of the farmer-operators across all villages and castes own the lands theyare cultivating; however, they suffer from a lack of an assured source of water for riceproduction. This causes delays in crop establishment and drought stress at some stagesof crop growth. In low-lying areas, the onset of heavy monsoon, accumulation ofrainwater, and slow and inadequate drainage cause delays in crop establishment anddamage to the standing crop from flooding. These conditions result in decreased riceyields as well as low overall farm productivity (Singh 1996). Farmers in the threevillages obtain supplementary irrigation through their own and rented tube wells (Table14). There is disparity by caste in terms of access to irrigation facilities. Of the totalfarming households in Chandpur, 23% and 13% of the upper and lower caste, respec-tively, invested in pump sets and tube wells. In Mungeshpur, a higher proportion(67%) was able to afford investing in supplementary irrigation facilities. Only 9% ofthe lower caste households in this village have their own tube wells and pump sets.Thus, they rent from the richer farmers. Farmers from the lower caste are often at themercy of upper caste households who have more access to water. Those who do notown irrigation facilities suffer from crop failure. Because of the high cost of supple-mentary irrigation, farmers in Mungeshpur use their own irrigation facilities not only

Table 14. Access to land and supplementary irrigation by village and caste (%),1995.


Farm characteristic Upper Lower Upper Lower Upper Lowercaste caste caste caste caste caste

(n = 13) (n = 138) (n = 12) (n = 121) (n = 6) (n = 75)

% of land (tenure status)Owned 88 81 98 94 100 97

Share tenant 12 9 2 6 – –Leasehold – 10 – – – 3

% of land irrigated byOwn tube well 21 40 66 17 34 34

Rented tube well 55 40 34 80 66 62Canal irrigation/ 24 20 – 3 – 4

none

% of farmers with 23 13 67 9 33 35own irrigationfacility


for rice but also for cash crops such as vegetables sometimes during the kharif andoften during the rabi season. Access to supplementary irrigation enables farmers togrow rice, wheat, berseem, and vegetables for consumption and for the market.

Poor farmers, particularly the widows who can’t afford to pay the rental fees ofwater for irrigation, exchange their labor for water but are often taken advantage ofby the tube well owners, who pay them lower than normal wage rates.

Human capitalA major factor influencing women’s productivity is the extent to which they haveaccess to education, training, and extension. There is general agreement that educa-tion increases productivity and a substantial literature exists documenting the posi-tive effects of women’s education on human capital development, paid labor forceparticipation, and agricultural production (Cloud 1985). There is a wide disparity inaccess to education by caste and gender of adult workers (Fig. 4A). Among all theadult females of the upper castes in Chandpur and Mungeshpur, about one-fourthhave not gone to school. Illiteracy rates among the lower caste female adults, how-ever, are very high at 80% in Chandpur and 91% in Mungeshpur. In both villages,most of the upper caste adult males are literate. In contrast, among the lower casteadult males, 37% in Chandpur and 58% in Mungeshpur were not able to go to school.Literacy rates among children 15 years old and below, especially for girls, are higherin Chandpur than in Mungeshpur among the lower caste (Fig. 4B). This trend is simi-lar in Sariyawan where female illiteracy is higher than male illiteracy.

Fig. 4. Literacy rates of (A) adult family members of farming house-holds (above age 16) and (B) males and females (15 years andbelow). UC = upper caste, LC = lower caste.

Mungeshpur

UC LCFemales

UC LCMales

100

80

60

40

20

0

B Chandpur

UC LCMales

UC LCFemales

Illiterate Literate

100

80

60

40

20

0

Literacy rate (%)

A Chandpur Mungeshpur

366 Paris et al

According to Bennett (1989), the higher illiteracy rates of women than men arecommon in eastern India. Five populous states (Andhra Pradesh, Bihar, MadhyaPradesh, Rajasthan, and Uttar Pradesh), wherein rice farming is predominantly rainfed,account for more than half of India’s illiterate females. These five states contain 89%of India’s districts where the rural literacy rate is below 5%, 83% of those with ratesof 5–9%, and 67% with between 10% and 14%. Illiteracy is widespread among ruralwomen of the scheduled castes and scheduled tribes.

The reasons for the low literacy rates among women and girls are social, cul-tural, and economic. Among poor farming and landless households, the need for childlabor within and outside the home is a major reason for boys and girls not to attend orto drop out of school. Girls are expected to help with the domestic chores, substitutefor their mothers in taking care of younger siblings, and help in field activities whileboys help tend animals after school. Another reason for the women’s lack of access toeducation is the greater limitation parents put on a girl’s freedom of movement, whichmay prevent her from going to school after a certain age. According to Mukhopadhyay(1984), girls who have reached the age of puberty are withdrawn from schools be-cause of the “social dangers” associated with male school teachers and students. Thus,socialization, gender roles, and sexual mores all play important roles in deprivinggirls of formal education (Bennett 1989). Girls are married off at an early age, thusconfining them to the status of daughter-in-law, which curtails their freedom of move-ment, association, and communication even further.

It is also traditionally believed that sons are more important because a daughterwill leave her mother’s home and join her husband’s family after marriage. Sons areexpected to take care of their parents in their old age (whole life) and after death,when they will perform the last rituals. Another factor contributing to low educationlevels for girls is the small return anticipated from girls’ schooling. While boys’ edu-cation is viewed as an investment in families’ socioeconomic status and as old-agesecurity for parents, girls are destined to be married into other families and henceyield no returns to their parents (Bennett 1992). Girls will be mothers and workers inoccupations that require little formal education. Investment in boys’ education is likelyto pay more dividends in the future in terms of increased chances of employment andconsequent support of the family. In the dowry system, males who have higher educa-tion can request a higher dowry price. In addition to these reasons, the direct costs ofeducation also deter families from sending their girls to school. Although school edu-cation in India is entirely free, expenses on books and learning materials, uniforms,and transport can be a heavy burden on poor families.

Access to agricultural-related informationA focused survey of male and female farmers from the lower castes in the three vil-lages was conducted to determine their sources of agricultural information (Table15). A majority of the men and women interviewed obtained their technical knowl-edge from their neighbors. In both villages, households of the same caste clustertogether and obtain information through socialization. This information indicates theimportance of social networks and kinship in disseminating information and identify-


Table 15. Access to agricultural information of males and females,1995.

Village/source Male % Female %

ChandpurNeighbors 18 60 13 65Extension staff 5 17 3 15Research institute 4 13 4 20Radio 3 10 3 15Television 6 20 3 15Spouse and relatives 12 40 6 30

MungeshpurNeighbors 23 76 15 75Extension staff 7 23 2 10Research institute 6 20 5 25Radio 2 6Television 2 6Spouse and relatives 15 50 3 10

SariyawanNeighbors 21 70 15 75Extension staff 10 33 2 10Research institute 8 26 3 15Radio 4 13 2 10Television 3 10 1 5Spouse and relatives 5 16 8 40No. of respondents 30 20

ing the key persons or “shining stars” who can serve as agents of change. A lowpercentage of the female respondents receive information from extension agents. Onereason for this is the general lack of female extension workers who can directly inter-act with women farmers. FAO (1991) revealed that women constitute a mere 0.59%of India’s agricultural extension workers. Since almost all extension workers are men,women’s roles and skills are usually overlooked, even in areas where they do most ofthe work.

Summary and conclusions

This chapter demonstrated the use of gender analysis as an analytical tool in charac-terizing and understanding farm-household systems in rainfed lowland rice villages.The analysis showed that, within farm households, there are gender-specific roles andresponsibilities and gender differences in access to resources that have to be consid-ered by scientists and extension and development workers. Although poor women inrainfed lowland rice environments play vital roles in sustaining food security andalleviating poverty, they face several constraints that limit their potential for increas-ing farm productivity. These barriers to productivity are a lack of access to education,

368 Paris et al

training, extension, new seeds suitable to their environments, animal fodder, and equip-ment to ease their workload and overcome drudgery in performing their farm tasks.One strategy for enhancing their productivity and income is to provide them withaccess to new seeds suitable to their specific adverse rainfed environments. Effortsare now being made under the System-wide Initiative on Farmer Participatory PlantBreeding and Gender Analysis to develop methodologies that involve both male andfemale farmers in rice variety development in rainfed environments (Paris et al 1998).Other potential research areas for enhancing women’s roles will be in producing ani-mal fodder within the cropping systems and developing agricultural/mechanical imple-ments/tools to reduce their drudgery, increase their labor efficiency, and explore waysto optimize the use of rice by-products and home-based technologies. Gender analy-sis will also be replicated to complete the socioeconomic and cultural characteriza-tion of major rainfed lowland rice environments in South and Southeast Asia and toprovide a gender-related database for policymakers in addressing gender issues inagriculture.

ReferencesAgarwal B. 1998. Disinherited peasants, disadvantaged workers: a gender perspective on land

and livelihood. Econ. Polit. Wkly. p 2-14.Bennett L. 1992. Women, poverty, and productivity in India. EDI Seminar Paper Number 43.

Washington, D.C. (USA): Economic Development Institute of the World Bank.Bennett L. 1989. Gender and poverty in India: issues and opportunities concerning women in

the Indian economy. Washington, D.C. (USA): World Bank. 153 p.Chen M. 1990. Coping with seasonality and drought. New Delhi (India): Sage Publications.

247 p.Cloud K. 1985. Women’s productivity in agricultural systems. In: Overholt C, Anderson MB,

Cloud K, Austin J, editors. Gender roles in development projects: a case book. WestHartford, Conn. (USA): Kumarian Press. p 57-78.

Dreze J, Sen A. 1989. Hunger and public action. Oxford (UK): Clarendon Press.Duvvury N. 1989. Women in agriculture: a review of the Indian literature. Econ. Polit. Wkly.

28 October 1989.FAO (Food and Agriculture Organization). 1991. Most farmers in India are women. New Delhi

(India): FAO. 20 p.Feldstein HS, Poats SV, Cloud K, Noreem R. 1989. Intra-household dynamics and farming

systems research and extension: conceptual framework and worksheets. In: FeldsteinHS, Poats SV, editors. Gender and agriculture: case studies in intra-household analysis.West Hartford, Conn. (USA): Kumarian Press.

Ghosh H. 1987. Changes in the status of north Indian women: a case study of Palitpur villages.Working Paper No. 141. Michigan State University, East Lansing, Mich. (USA).

Gopalan S. 1992. Sectoral policies on agriculture and related macro-economic policies andtheir gender-responsiveness. New Delhi, India. (In mimeo.)

Gupta AK. 1991. Reconceptualizing development and diffusion of technology for dry regions.In: Prasad C, Das P, editors. Extension strategies in rainfed agriculture. New Delhi (In-dia): India Society of Extension Education. p 322-356.


Hossain M. 1995. Recent developments in the Asian rice economy: challenges for rice re-search. In: Evenson R, Herdt R, Hossain M, editors. Rice research in Asia: progress andpriorities. Manila (Philippines): International Rice Research Institute and Center forAgriculture and Bioscience International. p 59-70.

IRRI (International Rice Research Institute). 1990. Gender analysis in rice farming systemsresearch: Does it make a difference? Report (unpublished) of the Women in Rice Farm-ing Workshop held in Indonesia, 4-8 June 1990.

IRRI (International Rice Research Institute). 1992. Proceedings of the international workshopon gender concerns in rice farming, Chiang Mai, Thailand, 22-25 October 1992. Manila(Philippines): IRRI.

IRRI (International Rice Research Institute). 1997. Sustaining food security beyond the year2000: a global partnership for rice research. Manila (Philippines): IRRI.

Kandiyoti D. 1991. Bargaining with patriarchy: social construction of gender. New Delhi (India):Sage Publications. p 114-118.

Kelkar G. 1992. Women, peasant organizations and land rights: a study from Bihar, India.Occasional paper. Gender and Development Studies, Asian Institute of Technology,Bangkok, Thailand. 50 p.

Mukhopadhyay M. 1984. Silver shackles: women and development in India. Oxford (UK):OxFam.

Norman D, Simmon EB, Hays M. 1983. Farming systems in the Nigerian Savanna: researchand strategies for development. Boulder, Col. (USA): Westview Press.

Pandey S, Singh HN, Villano R. 1998. Rainfed rice and risk-coping strategies: some macro-economic evidences from Eastern Uttar Pradesh. Paper presented at the NCAP-IRRIWorkshop on Risk Analysis and Management in Rainfed Rice Systems. 21-22 Sep 1998,New Delhi.

Paris T, Singh A, Hossain M, Luis J. 1998. Incorporating gender concerns in rice varietal im-provement and germplasm conservation: preliminary results in eastern Uttar Pradesh,India. Paper presented at the 2nd International Seminar of the CGIAR System-WideProgram on Participatory Research and Gender Analysis (SWP PRGA) held in Quito,Ecuador, 6-9 September 1998. (Forthcoming.)

Poats S. 1990. Gender issues in the CGIAR system: lessons and strategies from within. Paperpresented at the 1990 CGIAR Mid-Term Meeting, 21-25 May 1990, The Hague, TheNetherlands.

Quisumbing A. 1995. Women in agricultural systems. In: Strommoquist N, editor. Women inthe Third World: an encyclopedia of contemporary issues. New York (USA): GarlandPublishing, Inc. p 262-271.

Sarkarung S. 1996. Breeding rice cultivars suitable for rainfed lowland environmetns: a farmerparticipatory approach in eastern India. In: Eyzaguire P, Iwanaga M, editors. Participa-tory plant breeding. Proceedings of a Workshop on Participatory Plant Breeding, 26-29July 1995, Wageningen, The Netherlands. Rome (Italy): International Plant GeneticResources Institute.

Singh VP. 1992. Institutionalization of a farming systems approach to development. In: Pro-ceedings of technical discussions. Rome (Italy): Food and Agriculture Organization.

Singh VP. 1996. Monitoring and assessing the impact of a participatory research for the devel-opment of sustainable production systems: a decade of experience in rainfed situationsof Eastern India. Manila (Philippines): International Rice Research Institute.

370 Paris et al

Widawsky D, O’Toole JC. 1990. Prioritizing the rice research agenda for Eastern India. In:Evenson RE, Herdt RW, Hossain M, editors. Rice research in Asia: progress and priori-ties. Manila (Philippines): IRRI and CAB International.

NotesAuthors’ addresses: Thelma Paris, Affiliate Scientist-Gender Specialist, Social Sciences Divi-

sion, IRRI; Abha Singh, Sociologist, Kumarganj, Faizabad District, Eastern Uttar Pradesh;Mahabub Hossain, Agricultural Economist and Head, Social Sciences Division, IRRI;Joyce Luis, Assistant Scientist, Social Sciences Division, International Rice ResearchInstitute, DAPO Box 7777, Metro Manila, Philippines.

Acknowledgments: The authors wish to thank Dr. R.K. Singh, former Director of Research ofthe Narendra Deva University and Agricultural Technology (NDUAT) in Kumarganj,Faizabad District, Eastern Uttar Pradesh, and Dr. V.P. Singh, Agronomist, APPA Divi-sion, IRRI, for their suggestions. We are also grateful for the assistance in data analysisprovided by Ms. Josie Narcisco and Ms. Gemma Belarmino of IRRI.


Agricultural commercialization and land-use intensification: . . . 371

Low food security, high population growth, and environmental degradationare some of the major problems for agricultural intensification in the uplandsof Vietnam. Although Vietnam has now become a major rice exporter, foodproduction in these remote upland environments is insufficient to meet theneeds of the country’s growing population. To increase farmers’ incomesand their access to food, the government of Vietnam has encouraged produc-tion of cash crops in uplands through investments in marketing infrastruc-ture and institutional reform. Using farm-level survey data, this chapter ex-amines the effects of improved access to markets and improvements inlowland productivity on land-use intensity, labor productivity, and food secu-rity in the uplands. The results indicate that these changes have reduced theintensification pressure on uplands and generally improved the food securityof upland households.

Two-thirds of Vietnam’s natural area is classified as uplands, where 25 million people(one-third of the country’s population) are living. The uplands in Vietnam, as in otherparts of the developing world, are characterized by heterogeneous and fragile ecosys-tems, a high incidence of poverty, severe deforestation, and soil degradation. Theincreased population pressure caused by natural population growth as well as migra-tion of lowlanders has compounded these problems.

A major factor affecting the upland systems of Vietnam in recent years is a shiftin the outlook of the government toward these areas based on the recognition thatupland systems are an important component of the overall economy. As a result, thegovernment is undertaking additional investment to build rural infrastructure. In ad-dition, policy and institutional reforms are being undertaken to improve the welfareof people in these uplands. For example, new kinds of policies on property rightssuch as stewardship contracts are being promoted to encourage more sustainable useof land at the forest margin. Policies to discourage shifting cultivation and forestclearing for upland rice cultivation (MARD 1998) are being introduced. These efforts

Agricultural commercializationand land-use intensification:a microeconomic analysis of uplandsof northern VietnamNguyen Tri Khiem, S. Pandey, and Nguyen Huu Hong

372 Khiem et al

have improved market access in the uplands, encouraged a diversification of uplandproduction systems, and led to increasing commercialization of agriculture.

The process of commercialization and diversification of the upland systems hasalso been facilitated by rapid increases in the productivity of lowland areas. In the late1980s, Vietnam began a process of decollectivization, market reform, and trade liber-alization in combination with investment in water control and promotion of short-duration high-yielding rice varieties. These reforms stimulated a rapid expansion inrice production from 1986 to 1989 and Vietnam has now become the third largestrice-exporting country (Khiem and Pingali 1995, Minot and Goletti 1998). Improve-ments in food grain productivity in the lowlands have encouraged diversification inthe uplands as food needs are increasingly being met from the lowlands, thus releas-ing upland areas for more profitable uses, especially where marketing facilities arealso better.

These changes in the lowland rice economy and the impact of new policy ini-tiatives being undertaken to develop upland areas may increase or reduce the pressurefor intensification of upland system use. It is not possible to predict a priori the natureof adjustments in the upland systems that these changes might trigger. Some studieshave indicated that policy changes in the form of land allocation and more stable landtenure in the northern mountain region have led to an increase in crop yields andreforestation of formerly barren hills (Dovonan et al 1997, Tachibana et al 1998). Anincrease in lowland productivity can be expected to reduce the pressure for food pro-duction in the uplands (Coxhead and Jayasuriya 1994, Tachibana et al 1998). But itmay also increase land-use intensity by encouraging cash crop production and maylead to further exploitation of marginal lands (Barbier and Bergeron 1998, Hardakeret al 1993). There is evidence that as farmers in Vietnam have substituted high-yield-ing maize for upland rice in more favorable upland areas (Sikor and Truong 1998),upland rice production has been pushed to the more marginal land. In addition, inareas where marketing institutions are not well developed, a shift to nonfood cropscan increase the vulnerability of small farmers because of the uncertainty associatedwith the price of cash crops (Dewey 1981). On the other hand, commercial produc-tion can increase total household income and trigger further multiplier effects by en-couraging the adoption of improved technologies (von Braun and Kennedy 1994).

The manner in which upland households respond to improvements in marketaccess and to changes in economic opportunities is a critical factor in determining themerits of agricultural policies. The present study aims at examining the impact ofchanges in market access and population pressure on food supply at the householdlevel, cropping intensity in uplands, the extent of cash crop production, the composi-tion of household income, food security, and the productivity of land and labor. Theanalysis is based on an intensive survey of 980 households in 33 communes of 6provinces in the northern mountainous highlands during the crop years of 1997-98and 1998-99.


Conceptual framework

Population pressure has been considered to be a major factor leading to intensifica-tion of agriculture (Boserup 1965, 1981). Although the initial response to a popula-tion increase may be to expand the area, the closure of land frontiers ultimately willforce more intensive use of land as households attempt to satisfy their food needsfrom the shrinking land base per capita. This type of intensification will reduce laborproductivity as fallow periods are reduced and more and more labor is applied to agiven landholding. This situation can lead to a downward spiral of population growth-intensification-poverty unless technological change arrests a further decline in laborproductivity or a massive emigration reduces the population pressure.

Intensification may also result from an expansion of marketing opportunities.The additional demand created by marketing opportunities initially provides incen-tives for an expansion of area and ultimately for more intensified land use. Laborsupply is less likely to be a constraining factor as improved access to markets reducesthe cost of labor-substituting technologies such as mechanical tillage and herbicides.Better access to inputs such as fertilizers also improves the returns to land as well aslabor, thus further reinforcing the intensification process. The downward spiral re-sulting from population-driven intensification can be avoided when intensification ismarket-driven.

Theories of agricultural intensification and induced institutional innovation(Boserup 1965, Ruthenberg 1980, Hayami and Ruttan 1985) and agricultural house-hold behavior in the context of incomplete markets (de Janvry et al 1991) provideconceptual frameworks to study the impacts of commercialization (Pender et al 1998,von Braun et al 1991, Barbier 1998). The effects of commercialization and policymeasures on rural household income and consumption are mediated through complexrelationships. An expected increase in income and production capacity will motivatehouseholds to enter the exchange economy and become more commercialized. Themost important exogenous determinants of commercialization are population change,availability of new technologies, improved seeds or agronomic practices, investmentin infrastructure, macroeconomic policies, wages and employment opportunities, anddirect government action (Fig. 1). Household responses to population growth andcommercialization are reflected in their decisions on natural resource managementand household resource allocation, that is, mainly family labor, including intensifica-tion use of labor and capital per unit of land, reduction of fallow periods, adoption oftechnologies reflected in land-use patterns, and expenditures on food and nonfoodproducts.

Because of the complexity of modeling the interrelationships within rural house-holds and limited data, reduced-form models are used in this study to evaluate theeffects of population pressure and commercialization on land-use intensification, prod-uct choice, labor productivity, and food sufficiency.

374 Khiem et al

Recent trends of production systems in the northern uplands

Because of their rotational pattern and the difficulty in collecting information, esti-mates of cultivated uplands vary widely among the sources. The National Institute forAgriculture Planning and Projection (NIAPP 1993) estimated that, out of 2.7 millionha of cultivated land, about 1.4 million ha currently comprise swidden fields. If thefallow area is included, the total area under rice-swidden has been estimated to be 3.5million ha (Sam 1992) under shifting cultivation where upland rice is usually the firstcrop planted after slash-and-burn agriculture. Arraudeau and Xuan (1995) estimatedtotal upland rice area in Vietnam at 0.45 million ha with the total area under rice-swidden at approximately 8 million ha. According to the World Bank (1995), theupland rice system in the northern mountain region comprises mainly sedentary shift-ing cultivators who stay in one place but shift cultivation sites, affecting an area ofabout 1 million ha. Another 0.2 million ha is under itinerant cultivation practiced by afew ethnic groups.

Rapid population growth, driven by both high birth rates and in-migration oflowlanders, has brought about drastic changes in the economy of the northern up-lands. Poverty, environmental degradation, increased pressure on resources, and so-cial marginalization are interacting to create a downward spiral that is currently reachingcrisis proportions (Jamieson et al 1998). The northern uplands experienced an in-crease in population of more than 300% between 1960 and 1984. The incidence ofpoverty is much higher in the uplands than in other agroecological regions. The eightpoorest provinces are all located in the northern uplands (Minot 1998). The averageincidence of poverty in the northern mountain region is 31% compared with the na-

Fig. 1. Determinants and consequences of commercialization at thehousehold level.

Population,demographic change

Technologies,new crops

Infrastructure Prices,wages, risks

Household resource, endowment(land, labor, capital)

Resource allocation

Home goodsproduction

Off-farmwork

Agriculturalproduction

Commercialization effects

Marketedsurplus

Cash income

Foodconsumption

Nonfoodexpenditure

Subsistencefood

Macroeconomicpolicies, governmentinvestment, programs


tional average of 18%1, and most of these poor households are found among theethnic groups (MARD 1998).

Total food crop production in the region from 1985 to 1997 was barely ad-equate to meet the increased demand brought about by population growth. The percapita staple food crop output of the region remains at 250 kg despite a slight increasein rice, cassava, and sweet potato output. Investment in small-scale irrigation in somelowland fields helped expand wetland rice area by 1.3% per annum and yield in-creased 2% per annum from 1985 to 1997. The most remarkable change during themore recent years has been the expansion of area planted to improved maize or hy-brid maize from 150,000 to 250,000 ha (Fig. 2). There was a rapid increase in areaplanted to fruit trees and industrial crops; area growth rates of these crops were 8.3%and 2.5% per annum, respectively, from 1985 to 1997. The climate of the northernmountain region is suitable for a wide range of valuable fruits such as apricot, persim-mon, plum, jujube, tangerine, lychee, and others. A more favorable land tenure policyand improvement in access to markets have also encouraged farmers to plant fruit

1A poor household is defined as one having a monthly per capita income equivalent to 13 kg of milled rice orlower (about US$50 per annum).

Fig. 2. Index of areas planted to various crops in the northern moun-tain region (1985 = 100). Horticultural crops include vegetablesand beans. Industrial crops include sugarcane, groundnut, soybean,tea, coffee, and rubber.

170

160

150

140

130

120

110

100

90

80

701985 1987 1989 1991 1993 1995 1997

Year

227,000 ha

82,000 ha

59,000 ha

824,000 ha

RiceHorticultural crops and fruit treesIndustrial cropsMaize

376 Khiem et al

trees and other perennial crops. However, price fluctuations and unstable marketscaused by undeveloped postharvest facilities and marketing institutions are prevent-ing the stable development of fruit trees in the region. In many cases, farmers havebeen forced to cut down one type of tree in order to grow another tree species thatproduces more marketable products or they found the harvesting cost higher than themarket price of the product (Tuyen 1995, Dang 1993).

Although the area planted to upland rice is declining, it is the main food cropfor the millions of poor people and the ethnic minorities. Upland rice is grown aloneor in diverse mixtures in shifting or permanent fields under a wide range of condi-tions of climate, slope, and soil type. Upland rice area during 1980-85 expanded rap-idly because of food scarcity as the productivity of lowland rice was low. However,after the decollectivization of agriculture, the area of upland rice has declined, espe-cially since 1990. According to official statistics, the total area planted to upland ricein the northern mountain region is reported to be between 100,000 and 120,000 ha.Current state policy, however, is to discourage shifting cultivation and limit the areasopen for upland rice. Therefore, the reported upland rice area tends to be severelybiased downward. On the other hand, conducted surveys show that villagers failed toreport fields in remote areas or decreased the area under cultivation that is subject totaxation. The use of remote imagery in a selected commune in Son La Province foundthat the actual worked area is much larger than the reported area (Sikor and Truong1998).

Ethnicity and production systems in the northern mountainous region

The production systems in the northern uplands are generally determined by the dif-ferent agroecological conditions and cultural and food preferences of the diverse eth-nic groups. There are 31 ethnic groups belonging to seven language groups living inthe northern uplands. The six largest groups are the Tay (1 million), the Thai and theNung (0.6 million each), and the H’Mong, Muong, and Dzao (0.5 million each). Sev-enteen groups have populations under 10,000. Each group usually has its own dis-tinctive customs and traditions, socioeconomic characteristics, and community struc-tures. However, many ethnic groups are also quite diverse internally. For example,although sharing a common language, the H’Mong are divided into several distinctsubgroups (e.g., Red H’Mong, Black H’Mong, Flowered H’Mong). Many ethnicgroups live intermixed with one another within the same delimited territory. Of the109 districts and towns in the northern mountain provinces, 59 districts have ten ormore ethnic groups. Residential segregation by ethnic group is common only at thehamlet level (Khong Dien 1996, Vien 1997).

The H’Mong live on high slopes and mountain ridges, usually at altitudes above800 m. They cultivate maize or monocropped rice on swidden fields in combinationwith wet rice on terraces. The H’Mong of Meo Vac and Dong Van in Ha Giang Prov-ince inhabit high-altitude cold regions, and rely only on maize monocropping. Thesequence of crops after slash-and-burn agriculture commonly practiced by the H’Mongis sticky rice–nonsticky rice–maize–barley–Job’s tears (Coix lacryma) or cassava.


The Dzao live in medium-altitude areas; practice rotation of upland rice, maize,and cassava; and have experience in forestry and traditional garden crops. They prac-tice swidden cultivation by clearing fields from scrub vegetation and cultivating for 2to 4 years, followed by a fallow period of 10 years or more. They are specialists inconstructing terraces on upland slopes for paddy rice.

The Thai, Tay, Nung, and Muong live in low areas, intermontane valleys, andriver basins. They practice intensive wet rice agriculture. In growing food crops fortheir own consumption, the Thai prefer sticky rice, whereas the Tay and Nung prefernonsticky rice. The Nung, Tay, Giay, and others inhabit the lowland areas, at the baseof hills and in valley bottoms; paddy rice is their main crop. They have a remarkableirrigation technology for paddy production in terraced land; paddy rice output, how-ever, is not sufficient to satisfy food requirements and they need to plant root cropsand maize on slopes to supplement their diet.

Survey design and the data set

Using the a priori information on the upland systems of the northern mountain region,a stratified sampling design was carried out to generate a data set covering a widerange of population density, market access, ethnicity, relative proportion of uplandand lowland areas, and the extent of crop diversification. In the first step, 12 districtsin five mountain provinces were selected. The second step consisted of selecting ineach district two to three communes that differ in ethnicity and degree of marketaccess. Communes were classified as having good or poor market access by contrast-ing with other communes in the same district based on their relative degree of accessto the main transportation route (provincial and national roads) and district or provin-cial center markets in terms of physical distance in kilometers, accessibility by ve-hicles, and existence of a local market. In the final step, 30 to 50 households in eachcommune were randomly selected for structured interviews. In total, 980 householdsin 33 communes were included in the survey. Comparative analysis was carried out atthe household, commune, and district levels. Figure 3 shows the selected sites. Table1 shows the distribution of selected sites and their corresponding demographics, cropdiversification, physical distance to paved roads and district markets, and market ac-cess characteristics.

Table 2 summarizes the general characteristics of households in areas with poorand good market access. The average farm size is 1.72 and 1.32 ha for the two groupsof households, respectively. Households in locations with good market access aregenerally found at lower altitude, cultivate land with a lower slope, and have morelowland field area per capita. On average, households in these areas have 0.31 ha oflowland compared with 0.22 ha in the poor market access area. The extent of irriga-tion is greater in the low-slope areas. As a result of irrigation, cropping intensity oflowlands in these areas is higher than in lowlands of the upper slopes where only onecrop of rainfed rice per year is planted. In market-accessible areas, the major cashcrops are improved maize and horticultural crops. The proportionate area devoted tocash crops is usually higher in communes with better access to markets than other-

378 Khiem et al

Lai Chau

Lao Cal

Ha GiangCao Bang

Lang SoriTuyen Quang

Bac Thai

Ha Bac

Yen Bal

Sorr La

Hoa Binh

Vinh PhuPia Noi

Ha TaiHal Hung

Thai Binh

Quan Ninh

Nam HoNim Binh

Thanh Hoa

Nghe An

Ho Tinh

Fig. 3. Selected sites for socioeconomic studies in the northern moun-tain region.

wise. For example, 32% of the upland area is planted to cash crops (fruit trees, sugar-cane, beans, peanuts, medicinal crops) in areas with good market access, whereasthey occupy only 13% of the upland area in locations with poor market access.

Despite a smaller landholding, the rice output per household in areas with bet-ter access to markets is 32% higher, indicating a more favorable environment for riceproduction in these areas. A higher proportion of the lowland in total landholding, ahigher yield of lowland rice, and a higher intensity of rice production in the lowlandaccount for this difference. For areas with poor market access, households are muchmore dependent on upland rice. Their upland rice output accounts for about 50% oftotal rice production, while this percentage for households in areas with good marketaccess is only 16% (Table 2).

Cropping systems prevailing in the study area are extremely labor-intensivewhere upland rice, lowland rice, and maize are the main food crops, occupying morethan 70% of total cultivable uplands. Table 3 summarizes the total labor input perhectare for upland rice, lowland rice, and maize, which is further grouped into localand improved varieties. On average, farmers devote about 345, 344, and 250 person-days per hectare to the cultivation of upland rice, lowland rice, and maize, respec-tively. Land preparation, weeding, and harvesting account for almost 95% of theselabor inputs. Weeding labor occupies the largest share in the planting of upland riceand maize, and accounts for about 40% of total labor input.

Average returns to land and labor in the production of upland rice, lowland rice,local maize, and hybrid maize are also presented in Table 3. Returns to both land andlabor are highest for lowland rice and improved maize. Because of high returns to


Tabl

e 1

. S

elec

ted

dist

rict

s an

d co

mm

unes

and

sam

pled

hou

seho

lds.

Upl

and

Sam

pled

Popu

latio

nD

ista

nce

Dis

tanc

eD

omin

ant

Prov

ince

rice

area

Dis

tric

tco

mm

unes

dens

ityto

dis

tric

tto

Acce

ss b

yM

arke

tet

hnic

(ha)

(no.

of sa

mpl

ed(p

erso

nsm

arke

tpr

ovin

cial

vehi

cle

acce

ssgr

oup

hous

ehol

ds)

km

–2)

(km

)ro

ad (

km)

Cao

Ban

g3

,00

0N

guye

n B

inh

Tam

Kim

(3

9)

39

10

30

No

Low

Dza

oN

gan

Son

Thua

n M

ang

(38)

34

51

0Ye

sH

igh

Dza

o

Son

La

28

,00

0Th

uan

Cha

uC

hien

g Ph

a (1

3)

290

12

1Ye

sH

igh

Thai

Muo

ng G

iang

(1

2)

47

45

5Ye

s/no

aLo

wTh

aiC

o M

a (1

2)

34

50

50

No

Low

H’m

ong

Chi

eng

Kho

ang

(13)

143

30

20

Yes

Hig

hTh

ai

Mai

Son

Chi

eng

Mun

g (1

3)

164

15

2Ye

sH

igh

Thai

Chi

eng

Chu

ng (

13

)4

84

01

2Ye

s/no

aLo

wTh

aiTa

Hoc

(1

2)

39

30

20

No

Low

Thai

Phie

ng P

an (12)

27

70

15

No

Low

Thai

Bac

Yen

Ta X

ua (33)

19

810

No

Low

H’m

ong

Hon

g N

gai (

33)

38

810

No

Low

H’m

ong

Phie

n B

an (34)

77

24

Yes

Hig

hH

’mon

g

Ha

Gia

ng3

,00

0B

ac M

eYe

n C

uong

(33)

31

560

No

Low

H’m

ong

Yen

Phon

g (3

3)

50

56

0N

oLo

wD

zao

Yen

Phu

(34)

73

260

Yes

Hig

hTa

y

Yen

Bai

5,0

00

Mu

C. C

hai

Mo

De

(33)

50

12

7N

oLo

wH

’mon

gK

im N

oi (34)

43

10

5N

oLo

wH

’mon

gC

he C

u N

ha (

34

)46

62

Yes

Low

H’m

ong

Tram

Tau

Tram

Tau

(33)

44

220

No

Low

H’m

ong

Ban

Con

g (3

3)

15

520

No

Low

H’m

ong

Hat

Luu

(34)

202

15

2Ye

sH

igh

Thai

cont

inue

d on

nex

t pa

ge

380 Khiem et al

Tabl

e 1

con

tinu

ed.

Upl

and

Sam

pled

Popu

latio

nD

ista

nce

Dis

tanc

eD

omin

ant

Prov

ince

rice

area

Dis

tric

tco

mm

unes

dens

ityto

dis

tric

tto

Acce

ss b

yM

arke

tet

hnic

(ha)

(no.

of sa

mpl

ed(p

erso

nsm

arke

tpr

ovin

cial

vehi

cle

acce

ssgr

oup

hous

ehol

ds)

km

–2)

(km

)ro

ad (

km)

Lao

Cai

5,0

00

Bac

Ha

Bao

Nha

i (5

0)

76

15

20

No

Low

Dza

oN

a H

oi (50)

215

55

Yes

Hig

hTa

y

Bao

Yen

Yen

Son

(3

0)

10

11

83

Yes

Hig

hD

zao

Vinh

Yen

(40)

62

32

Yes

Hig

hTa

yN

ghia

Do

(30

)1

16

15

1Ye

sH

igh

Tay

Lai C

hau

25

,00

0Ph

ong

Tho

La N

hi T

han

(30)

20

91

2N

oLo

wD

zao

Sun

g Ph

ai (30)

53

58

Yes/

noa

Hig

hH

’mon

gN

am L

oong

(40)

48

610

No

Low

H’m

ong

Tuan

Gia

oQ

uai N

ua (33)

72

10

1Ye

sH

igh

Thai

Pu N

hung

(33)

47

15

6N

oLo

wH

’mon

gPh

ing

San

g (3

4)

44

25

20

No

Low

H’m

ong

a Acc

essi

ble

only

by

two-

whe

el v

ehic

les

up t

o vi

llage

cen

ter.


Table 2. General characteristics of the sampled households (1998-99).

Item Poor market access Good market access

Number of sampled households 598 382Average household size 7.84 (0.15)a 6.81 (0.13)Annual average per household (kg)

Rice production 1,323 (45.34) 1,533 (52.29)Lowland rice 660 (37.18) 1,289 (50.13)Upland rice 663 (24.12) 244 (15.43)

Rice purchase 275 (13.99) 248 (16.97)Average distance from markets (km) 30 10Per capita rice supplyb (kg) 210 (4.42) 261 (5.76)Average farm size (ha) 1.72 (0.07) 1.32 (0.08)

Lowland 0.22 (0.02) 0.31 (0.01)Upland 1.50 (0.06) 1.01 (0.07)

Lowland rice area/lowland area 1.14 1.40Upland rice area/upland area 0.42 0.27Upland maize area/upland area 0.29 0.26Garden and orchard area/upland area 0.13 0.32

aNumbers in parentheses are standard errors. bSum of rice production and net purchase.

Table 3. Total labor use, yield, and returns to land and labor for up-land rice, lowland rice, and maize.

Item Upland Lowland Local Improved orrice rice maize hybrid maize

Sample size 697 672 507 207Total person-days ha–1 345 344 250 281Labor time allocation (%)

Land preparation 27 26 39 27Seeding/crop 9 19 8 11

establishmentWeeding 42 16 30 35Fertilizer application 1 8 – 2Pest control – 1 – –Irrigation – 9 – –Harvesting 20 21 22 26

Yield (t ha–1) 1.56 3.15 1.50 3.30Average returns to

Land ($ ha–1) 197 393 133 282Labor ($ person-day–1) 0.57 1.10 0.65 1.00

382 Khiem et al

land, lowland fields are almost invariably planted to rice. Although returns to land arehigher for upland rice than for local maize, the high labor requirement for upland rice,especially for weeding, lowers the returns for labor. Both labor and land productivityare higher for hybrid maize than for upland rice. The higher productivity of hybridmaize has induced a rapid expansion of its area.

Upland rice is usually the crop planted in the first one or two years after slash-and-burn agriculture, followed by maize and cassava when the soil becomes less fer-tile. While wet rice from lowland fields becomes the main source of rice supply,farmers usually maintain supplementary upland rice on swidden fields, mostly stickyrice, to satisfy household needs. In the more favorable upland environment with goodmarket access and where farmers have sufficient access to lowland rice cultivation,there is an increasing trend toward substitution of hybrid or improved open-polli-nated variety maize for upland rice. Most of the hybrid maize output is used in thehousehold as feed for domestic animals and a small portion is sold to markets.

Rice is the main food crop of the uplands. Although production of maize, espe-cially improved maize, is mainly used for domestic animals, some ethnic minoritiesstill depend on maize as their main diet. Other staples such as cassava, sweet potato,and canna (Canna edulis, a root crop) are consumed when rice and maize are in shortsupply. Most of the households in the study areas reported that they were unable tomeet the family food requirement from their own production of rice and maize. Onaverage, 88% of rice and maize consumption is from own production. This propor-tion of subsistence food production is almost equal among the two groups of house-holds in the poor and good market access areas.

The frequencies of reported food shortage are presented in Figure 4. On aver-age, the extent of food shortage is greater in areas with poor access to markets (2.4mo) than in areas with good market access (2.1 mo). The chi-square test for the differ-ence in the two distributions gave a value of 37.17 (9 degrees of freedom), which isstatistically significant at 1%. It is expected that households in areas with good mar-ket access having more diversified sources of income would depend more on the

Fig. 4. Distribution of households reporting foodshortage by number of months of food short-age in 1997-98.

35

30

25

20

15

10

5

0

Percentage of households

0 1 2 3 4 5 6 7 8Number of months of food shortage

9

Poor market accessGood market access


markets for their food needs. With the marketing system still being undeveloped,however, the day-to-day concern of households is how to produce sufficient food.Figure 5 presents the relationship between the number of reported months of foodshortage in the crop year 1997-98 and the average landholding per capita of the house-holds reporting a food shortage. The incidence of food shortage seems to depend onthe per capita availability of both uplands and lowlands. Households that reported afood shortage in more than 6 mo a year have half the size of both lowland fields andupland fields in comparison with those that suffered a food shortage only 1 mo ornone (Table 4). About 5% and 8% of the sampled households reported having morethan 6 mo of food shortage in poor and good market access areas, respectively. The

30

25

20

15

10

5

00 1 2 3 4 5 6 7 8

Percentage of households

1,600

1,400

1,200

1,000

800

600

400

200

0

Landholding(m2 capita–1)

Number of months of food shortage

Percentage of householdsUpland holdingsLowland holdings

Fig. 5. Relationship between the number of re-ported months of food shortage in 1998 and theaverage of landholding per capita of the house-holds reporting a food shortage.

Table 4. Landholding, annual cash income, and rice purchase of households facing food short-age of selected frequencies in 1997-98.

Poor market access Good market access

Months of reported More than 1 mo or More than 1 mo orfood shortage 6 mo none 6 mo none

Percentage of householdsa 5 32 8 46Upland area (ha) 0.56 1.22 0.29 0.53Lowland area (ha) 0.14 0.27 0.15 0.37Cash income (US$ household–1) 69 192 383 276Rice production (kg household–1) 506 90 491 63Rice purchase (kg household–1) 706 1,829 508 2,112

aThe percentages are calculated separately for the two groups of households, 598 in poor market access areasand 382 in good market access areas. Only households that reported a food shortage in more than 6 mo or inonly 1 mo or none during crop year 1997-98 are included in the table.

384 Khiem et al

Table 5. Cash and noncash income per household and sourceof cash income by market access.

Poor GoodItem market market access

access

Total income (US$) 517 638Share of cash income (%) 35 45Source of cash income (US$)

Animals 72 128Rice 4 4Maize 5 13Other crops 29 30Off-farm work 28 74Forest products 6 6Garden and orchard products 4 29

Noncash income (US$) 369 354

cash income of this group of households in poor market access areas, however, ismuch lower than that of their counterparts in the good market areas ($69 versus $383).They are the most food-insecure group of upland households.

The average cash income derived from sales of home garden and orchard prod-ucts of households in the good market access areas is seven times higher than that ofhouseholds in the poor market access areas (Table 5). The former’s total cash incomeis about 1.5 times higher, whereas noncash income of the two groups is almost equal.Overall, total annual income per household averaged at the district level varies from$340 in Bac Me District to $800 in Thuan Chau District. Contrary to the widely heldbelief that the upland system in the northern mountain region is highly subsistence-oriented, income derived from exchange with markets contributes to 30–60% of totalhousehold income. The sale of domestic animals is the highest single source of cashincome, contributing to about 40% of cash income for both groups of households.

Hypotheses and model specification

Rice is the major staple crop in Vietnam. For household consumption purposes, ricefrom lowlands and uplands is an almost perfect substitute. Thus, an improvement inthe productivity of lowland paddy can be expected to reduce the pressure for intensi-fication of food production in the uplands. Empirical testing of this hypothesis isdone by parameter estimation of a reduced-form model in which intensification isspecified as a function of population density, the proportion of lowland area, the aver-age slope of upland fields, and market access. The model is estimated by ordinaryleast squares. Following Ruthenberg (1980), the intensification index is calculated asthe ratio of growing period in years to the sum of the growing period and the fallowperiod. Farmers responsed to rising population density and unavailable land for fur-


ther “extensification” in the uplands by reducing fallow periods and prolonging thecropping cycles of food crop production, which resulted in a higher intensificationindex. Thus, in calculating the index, the land used for annual food crop productiononly is considered. Upland fields with higher slope can be considered to be of poorerquality due to their greater susceptibility to soil erosion. Dummy variables for differ-ent ethnic groups were specified to examine the effect of ethnicity on the dependentvariable. In the model, a positive coefficient of the population density variable wouldsupport the hypothesis of population pressure-driven intensification. The expectedsign of the ratio of lowland to total landholding is negative as food production in thelowlands substitutes for food production in the uplands. The coefficient of slope isexpected to be negative due to the land quality constraint to intensification. An im-proved access to markets is expected to reduce the pressure to intensify uplands forfood production as income generated from commercial crops can be used to purchasefood.

The second model attempts to explain the variations in the importance of up-land rice across households. The importance of upland rice is measured by the ratio ofupland rice area to upland area. Since upland rice is mainly a subsistence crop, thisratio is also a proxy for the degree of subsistence orientation in the use of uplands.Other variables are as for the first regression. As with the first regression, populationdensity is expected to have a positive effect and the proportion of lowland area andbetter access to markets are expected to have negative effects on the dependent vari-able. Since the dependent variable is censored, the model is estimated by Tobit re-gression.

In the third model, labor productivity in upland agriculture is specified to de-pend on farm size, household capital, labor endowments, and market access. Exceptfor the market dummy, all the variables in the model are expressed in logarithm. Themodel is estimated using ordinary least squares. Following the Boserupian argument,an increase in population pressure (or a reduction in farm size per capita) is expectedto reduce labor productivity, ceteris paribus. Thus, the expected sign of the coeffi-cient of farm size, which is a proxy for population pressure, is positive. Householdcapital endowment is also expected to have a positive effect on labor productivity.Improved market access is similarly expected to have a positive effect by encourag-ing a switch to high-value cash crops.

The fourth model attempts to determine the factors explaining the frequenciesof food shortage, which is defined as the number of months in a year that the house-hold encountered a short supply of food from its own production. The number ofmonths of food shortage in any particular year depends on several factors (Pandeyand Minh 1998). First, production may not be adequate due to adverse weather condi-tions. Second, the incidence of food shortage depends on family size, which deter-mines both demand for food as well as labor-supplying capacity. The demand forfood grain depends on total family size, whereas labor-supplying capacity dependson the number of adult family members. Thus, the incidence of food shortage is likelyto be higher among households that have a lower proportion of adults. Third, the area

386 Khiem et al

Table 6. Estimation of the models.a

Land-use Proportion Labor ReportedVariable intensity of upland productivity months of

index rice area in uplands food shortage

Sample size 710 980 960 980Constant 0.71 0.36 1.08 0.77Population density 0.14** 0.04 1.37**Household size 0.08**Farm size 0.32**Upland area –0.36**Lowland area –1.58**Proportion of lowland –0.20** –0.32**Average upland slope –0.12 –0.16Capital 0.13**Market access dummy –0.17** –0.01 0.36** –1.00**H’mong ethnic dummy –0.23** 0.13** 0.81**Dao ethnic dummy –0.09** 0.22** 0.85**Thai ethnic dummy –0.14** 0.20** 0.26R-square 0.29 0.12Pseudo R-square 0.16 0.14F-value 40.08** 40.93**Log-likelihood function –623.2 –1,890.7% censored 20.6 27.1

aModels 1 and 3 are estimated by ordinary least squares. Models 2 and 4 are estimated byTobit regression. *, ** denote statistical significance at 5% and 1% level, respectively. Thedependent variables are Model 1: Land-use intensity index defined as number of years of crop-ping of food crops divided by the sum of years of cropping of food crops and years of fallowperiod. Model 2: Proportion of upland area planted to upland rice. Model 3: Logarithm of grossreturns per day in thousand dong of family labor used for agricultural and home goods produc-tion. Model 4: Months of reported food shortage in 1997-98. Variables are defined as popula-tion density = population density of communes in number of persons per ha; household size =number of persons; farm size = logarithm of farm size measured in ha; area of lowland and areaof upland are measured in ha; proportion of lowland = proportion of lowland area in totallandholding; average upland slope = average of slope of all upland parcels cultivated by thehousehold measured in hundredth degrees; market access = dummy variable, poor marketaccess = 0, good market access = 1; capital = logarithm of total value of farm tools and animalstock per adult person in million dong; base ethnic dummy = the Tay.

of land operated and soil quality determine production potential. Households withsmaller land areas or poorer soils are more likely to suffer from food shortage. Sincethe dependent variable is censored, the model is estimated by Tobit regression.

Model results

The regression results are presented in Table 6. In the first model, all coefficientsexcept that of slope are statistically significant. The positive coefficient associatedwith population density supports the Boserupian hypothesis. The results also indicatethat improved access to markets and increased productivity of lowlands can reducethe pressure for intensification of uplands. Thus, investments in research to improve


the productivity of lowland rice can have a positive environmental impact by reduc-ing the incentive to intensify food production in the fragile uplands. Improvements inaccess to markets can generate a similar effect through income enhancement. Nega-tive statistically significant coefficients of ethnic dummies reflect location-specificeffects. Compared with the H’Mong, Dao, and Thai, the Tay, who live in lower areasand are used as a base dummy, practice more intensive cropping on their upland plots.

In the second model, the proportion of lowland area and the ethnic variables arefound to have a statistically significant effect in determining the proportionate areaallocated to upland rice. This result supports the hypothesis that an improvement inthe productivity of lowland paddy will reduce pressure to produce rice in the uplands.In doing so, it will free up uplands for other crops that may be cash crops. Whethersuch a switch will have a positive environmental effect is uncertain since it will de-pend on several other factors such as the types of cash crops chosen and the institu-tional arrangements for growing such crops. If upland rice is substituted by moreerosive annual cash crops, the environmental effect could be more detrimental. Infact, this seems to be the case in some parts of Vietnam where upland rice is beingreplaced by maize. As farmers plow the fields more intensively for maize than forupland rice, soil erosion problems in the maize fields, which also have poorer canopycover, may increase.

The third model indicates that labor productivity in upland agriculture is posi-tively related to farm size (which is inversely related to population pressure). Theparameter estimates suggest that an increase in population pressure (i.e., a reductionin farm size per capita) by 1% will reduce labor productivity by 0.32%. Thus, at thecurrent rate of population growth of 3% per year in the uplands of Vietnam, laborproductivity will decline by 1% per year. Thus, productivity improvement at the rateof 1% per year is needed just to maintain labor productivity. The signs of all othervariables are as expected.

In the fourth model, access to markets and farm size are found to have statisti-cally significant negative effects on the frequencies of food shortage. A higher abso-lute value of the coefficient associated with the size of lowland holdings reflects thehigher food productivity of the lowlands in food production. Investment in improve-ment of lowland rice productivity therefore helps enhance the status of food securityof upland households. Population density as a proxy for land scarcity and family sizehas a significant positive effect on the frequency of food shortage. The incidence offood shortage depends on family size, which determines both demand for food aswell as labor-supplying capacity. The positive and statistically significant effect offamily size on incidence of food shortage implies a stronger demand effect for foodgrain than the labor-supplying capacity effect.

Conclusions and policy implications

Improvement in market access and policy reforms for land tenure affect the alloca-tion of land-use patterns in the northern uplands. Secondary data show that areasplanted to fruit trees and horticultural crops have increased dramatically in the past

388 Khiem et al

ten years, whereas the increase in area planted to rice and other staple food crops hasbeen just enough to maintain the region’s level of per capita food output.

The results of the study indicate the important role that access to markets canplay in arresting and reversing the Boserupian decline in labor productivity as popu-lation-driven intensification of land use occurs. Both land and labor productivity werehigher in areas with better market access. The improvement in productivity and in-come resulted mostly from an expansion in cash crop production. Improvements inmarket access also reduced the need to intensify food production in the uplands. Thesefactors may have also generated positive environmental benefits.

These positive effects of market access, however, probably would not havematerialized unless the productivity of lowlands increased to improve the food secu-rity of farmers. Unlike in other countries, Vietnamese upland farmers also have somelowland fields in river basins and intermontane valleys. A rapid improvement in theproductivity of lowland fields as a result of policy changes during the mid-1980s wasinstrumental in relaxing food supply constraints to the diversification of land use inthe uplands. In fact, upland rice area, which had expanded rapidly during the 1970swhen lowland rice productivity was low, started to decline after the mid-1980s. Im-provements in market access alone without changes in the productivity of rice in thelowlands would probably not have resulted in more commercial production in theuplands.

These results highlight the importance of taking measures to assure food secu-rity as a prerequisite for a move toward more commercialized production systems inthe Asian uplands. In other countries and regions where farmers do not have access tolowlands to secure their food supplies, additional food production must come eitherfrom improvements in the productivity of uplands or through stable marketing chan-nels. Since marketing institutions in upland areas of most Asian countries are poorlydeveloped, improvements in the yield of food crops in uplands through agriculturalresearch and policy reform are needed to encourage changes in land use toward in-come-generating activities.

ReferencesArraudeau M, Xuan VT. 1995. Opportunities for upland rice research in Vietnam. In: Denning

G, Xuan VT, editors. Vietnam and IRRI: a partnership in rice research. Manila (Philip-pines): International Rice Research Institute and Hanoi (Vietnam): Ministry of Agricul-ture and Food Industry. p 191-198.

Barbier B. 1998. Impact of market and population pressure on production, incomes and naturalresources in the dryland savannas of West Africa: bioeconomic modeling at the villagelevel. EPTD Discussion Paper No. 21. Washington, D.C. (USA): International FoodPolicy Research Institute.

Barbier B, Bergeron G. 1998. Natural resource management in the hillsides of Honduras:bioeconomic modeling at the micro-watershed level. EPTD Discussion Paper No. 32.Washington, D.C. (USA): International Food Policy Research Institute.

Boserup E. 1965. The conditions of agricultural growth. New York (USA): Aldine PublishingCo.


Boserup E. 1981. Population and technology. Oxford (UK): Blackwell.Coxhead I, Jayasuriya S. 1994. Technical change in agriculture and land degradation in devel-

oping countries. Land Econ. 70:20-37.Dang BV. 1993. Economic and cultural changes in the northern mountain provinces. Hanoi

(Vietnam): Social Science Publisher.de Janvry A, Fafchamps M, Sadoulet E. 1991. Peasant household behavior with missing mar-

ket: some paradoxes explained. Econ. J. 101:1400-1417.Dewey KG. 1981. Nutritional consequences of the transformation from subsistence to com-

mercial agriculture in Tabasco, Mexico. Human Ecol. 9:151-187.Dovonan D, Rambo AT, Fox J, Cuc LT, Vien TD, editors. 1997. Development trends in Vietnam’s

northern mountain region. Hanoi (Vietnam): East-West Center and Center for NaturalResources and Environmental Studies.

Hardaker JB, Fleming E, Tin HN. 1993. Economic aspects of environmentally endangeredupland farming systems in the Asia-Pacific Region. In: Proceedings of the Workshop onUpland Agriculture in Asia, 6-8 April 1993, Bogor, Indonesia.

Hayami Y, Ruttan VW. 1985. Agricultural development: an international perspective. Balti-more, Md. (USA): The Johns Hopkins University Press.

Jamieson N, Cuc LT, Rambo AT. 1998. The development crisis in Vietnam’s mountains. Hono-lulu, Haw. (USA): East-West Center Special Report (6).

Khiem NT, Pingali PL. 1995. Supply responses of rice and three food crops in Vietnam. In:Denning G, Xuan VT, editors. Vietnam and IRRI: a partnership in rice research. Manila(Philippines): International Rice Research Institute and Hanoi (Vietnam): Ministry ofAgriculture and Food Industry. p 275-290.

Khong Dien. 1996. Socio-economic characters of the ethnic minorities in the northern moun-tain region. Hanoi (Vietnam): Social Science Publishing House. (In Vietnamese.)

MARD (Ministry of Agriculture and Rural Development). 1998. Socioeconomic developmentstrategies for the Northern Mountain Region 2000-2010. Hanoi (Vietnam): MARD.35 p.

Minot N. 1998. Generating disaggregated poverty map: an application to Vietnam. MSSDDiscussion Paper No. 25. Washington, D.C. (USA): International Food Policy ResearchInstitute.

Minot N, Goletti F. 1998. Export liberalization and household welfare: the case of rice in Viet-nam. Am. J. Agric. Econ. 80:738-749.

NIAPP (National Institute for Agricultural Planning and Projections). 1993. Bare lands in Viet-nam. Hanoi (Vietnam): NIAPP.

Pandey S, Minh DV. 1998. A socio-economic analysis of rice production systems in the up-lands of northern Vietnam. Agric. Ecosyst. Environ. 1373:1-10.

Pender J, Place F, Ehui S. 1998. Strategies for sustainable agricultural development in the EastAfrican highlands. Paper presented at the International Conference on Strategies forPoverty Alleviation and Sustainable Resource Management in the Fragile Lands of Sub-Saharan Africa. Entebbe, Uganda.

Ruthenberg H. 1980. Farming systems in the tropics. 3rd edition. Clarendon (UK): OxfordPress.

Sam DD. 1992. National background paper on shifting agriculture in Vietnam presented at theWorkshop on “Shifting agriculture in Laos and Vietnam, its social, economic and envi-ronment values to alternative land uses,” Chiang Mai, Thailand, August 1992.

390 Khiem et al

Sikor T, Truong DM. 1998. Sticky rice, collective fields: community-based development amongthe Black Thai. Hanoi (Vietnam): Center for Natural Resources and Environmental Stud-ies.

Tachibana T, Trung NM, Otsuka K. 1998. From deforestation to reforestation through tenurereforms: the case of the Northern Hill region of Vietnam.

Tuyen BC. 1995. Community-based natural resources management in Lam Dong province. In:Rambo AT et al, editors. The challenges of highland development in Vietnam. Honolulu,Haw. (USA): East-West Center.

Vien TD. 1997. Ethnic culture and farming systems in northern Viet Nam. UNESCO Work-shop on “Cultural aspects of natural resources management.” Hanoi, Vietnam.

von Braun J, de Haen H, Blanken J. 1991. Commercialization of agriculture under populationpressure: effects on production, consumption, and nutrition in Rwanda. Research Re-port No. 85. Washington, D.C. (USA): International Food Policy Research Institute.

von Braun J, Kennedy E. 1994. Agricultural commercialization, economic development andnutrition. Baltimore, Md. (USA): Johns Hopkins University Press.

World Bank. 1995. Vietnam: environmental program and policy priorities for a socialist economyin transition. Agricultural and Environment Operations Division. Washington, D.C.(USA): The World Bank.

NotesAuthors’ addresses: Nguyen Tri Khiem, Can Tho University, Vietnam, and Social Sciences

Division, International Rice Research Institute; S. Pandey, Social Sciences Division,International Rice Research Institute; Nguyen Huu Hong, Thai Nguyen University, Viet-nam.

Acknowledgments: The authors wish to acknowledge comments and generous support fromProfessor Hermann Weibel of Hanover University.


Economics of intensive rainfed lowland rice-based cropping systems . . . 391

Farming in the rainfed lowlands of Ilocos Norte, Philippines, is highly inten-sive, diversified, and commercialized. The cropping system is predominantlyrice-based in the wet season and high-value cash crops are grown during thedry season. The profitability of cash crop production has encouraged farm-ers to use high levels of purchased inputs such as chemical fertilizers andpesticides. Concerns are being raised about the long-run sustainability ofsuch intensive systems. This chapter assesses the sustainability of suchsystems using a total factor productivity analysis.

The trend in total factor productivity was positive (1992-95) initially butthen became negative (1996-97). However, the total factor productivity esti-mates for six years do not show any clear negative trend. Groundwater pollu-tion, particularly with NO3-N, has occurred as a result of the excessive use offertilizers on dry-season crops. If the effect of this negative externality werealso captured in the total factor productivity estimates, the decline for recentyears could have been sharper. Although fluctuations in total factor produc-tivity within the short period analyzed may be attributed mainly to climaticfactors, technological interventions for improving input-use efficiency arenevertheless needed to reduce both the input cost and the contamination ofgroundwater.

Farming in the rainfed lowlands of northwest Luzon, Philippines, is highly intensive,diversified, and commercialized. As in other rainfed environments, weather condi-tions are erratic with interspersed dry spells. The cropping system is predominantlyrice-based in the wet season (WS) from May to October and high-value crops aregrown during the dry season (DS) from November to April. Garlic (Allium sativumL.), maize (Zea mays L.), mungbean (Vigna radiata L. Wilczek), sweet pepper (Cap-sicum annum L. var. annum), and tomato (Lycopersicon esculentum L.) are amongthe most common DS crops grown after rice (Fig. 1). Some farmers maximize land-use intensity by growing two or three cash crops. These crops depend on irrigationwater from tube wells.

Economics of intensive rainfed lowlandrice-based cropping systemsin northwest Luzon, PhilippinesM.P. Lucas, S. Pandey, R.A. Villano, D.R. Culannay, and T.F. Marcos

392 Lucas et al

The rainfed lowlands of Ilocos Norte demonstrate a case of high cropping in-tensity and input use and can serve as a model for other rainfed areas that are beingintensified. Farmers in Ilocos Norte use high levels of fertilizer and pesticides be-cause of the economic benefits derived from high-value crops. These high-input in-tensive systems, however, may be unsustainable in the long run. This chapter presentsan economic analysis of intensive rice-based production systems in Ilocos Norte, thenorthernmost province in the Philippines. The possible sustainability or unsustainabilityof the system is also addressed.

The province of Ilocos Norte has a total land area of 362,000 ha, of which 25%is devoted to agriculture and forestry. There are 39,000 ha of rainfed areas, of which66% is rainfed lowlands. Agriculture and forestry directly employ 47% of the laborforce. In 1998, the province reached a sufficiency level of 256% for rice, 145% foryellow maize, 124% for root crops, 2,277% for garlic, 3,463% for onion, and 170%for mungbean. In 1998, rice area declined slightly due to natural calamities such as ElNiño and La Niña and two super typhoons that passed through the province. In spiteof these, however, the province still produced a surplus of rice. The cropping inten-sity increased from 130% in 1995 to 151% in 1998. The cropping intensity is ex-pected to increase further to 180% by 2001. This increase is expected to be achievedmainly through irrigation development projects.

Trends in area planted and yield of rice and nonrice crops

The area planted to rainfed rice in Ilocos Norte decreased in the past three years (Fig.2), whereas the area planted to irrigated rice has generally increased. The decrease in

Fig. 1. Cropping calendar and average monthly rainfall in Ilocos Norte,Philippines. Source: MMSU-PAGASA-PCARRD Agromet Station,MMSU Batac, Ilocos Norte, 1976-98.

600

500

400

300

200

100

0

Rainfall (mm)

Crops planted after

Other

Tobacco

Onion

Sweet pepper

Tomato

Mungbean

Maize

GarlicRice

May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May

Month


area under rainfed rice was attributed to the prolonged drought experienced in theprovince. The area under irrigated rice increased, however, with the expansion ofirrigation facilities. The yield of irrigated rice increased rapidly with a growth rate of6.3% per year, which was higher than that of rainfed rice (Fig. 3).

The area planted to dry-season crops in the province varied widely over time(Fig. 4). Maize area increased rapidly between 1994 and 1996 and stabilized. Garlicand mungbean areas increased from 1996 to 1998, while the area planted to sweetpepper remained highly variable. Tomato area remained stable because most of thefarmers are contract growers obtaining a fixed quota from the National Food Corpo-ration tomato paste processing plant. Except for tomato and sweet pepper, yields ofdry-season crops did not vary much over time (Fig. 5).

Fig. 2. Area planted to rice in 1991-98, IlocosNorte, Philippines. Source: Bureau of AgriculturalStatistics (1998).

Fig. 3. Average yield of rice in 1991-98, IlocosNorte, Philippines. Source: Bureau of AgriculturalStatistics (1998).

50,000

40,000

30,000

20,000

10,000

0

Irrigated

Rainfed

19911992

19931994

19951996

19971998

Year

Area (ha)

4.0

3.5

3.0

2.5

2.0

1.5

0

Irrigated

Rainfed

19911992

19931994

19951996

19971998

Year

Yield (t ha–1)

394 Lucas et al

Fig. 4. Index of area planted to cash crops inIlocos Norte, Philippines, 1991-98.

Fig. 5. Average yield of major dry-season cropsin rainfed areas in Ilocos Norte, Philippines,1991-98. Source: Bureau of Agricultural Statis-tics (1998).

Methodology

Study area and sampling designIlocos Norte is divided into four major regions: northern coastal, central lowlands,southern coastal, and eastern interior (PPDO 1995). Most of the agricultural activitiesare in the central lowlands composed of ten municipalities: Bacarra, Laoag City, SanNicolas, Dingras, Batac, Paoay, Sarrat, Currimao, Badoc, and Pinili. The monitoringof rice and nonrice production practices in farmers’ fields began in the 1991 WS andcontinued until the 1998 DS. Panel data from 100 randomly selected farmers werecollected.

350

300

250

200

150

100

50

01991

19921993

19941995

19961997

1998Year

Area index

MaizeGarlicMungbeanSweet pepperTomato

1816141210

86420

19911992

19931994

19951996

19971998

Year

Yield (t ha–1)

MaizeGarlicMungbeanSweet pepperTomato


The economic characterization and initial sustainability analysis were alreadypresented in an earlier publication (Lucas et al 1999). This chapter updates the earlierpaper by including more recent data that cover the period from the 1994 WS to 1998DS. The analysis on the possible sustainability or unsustainability of the croppingsystem, however, is based on the data for 1992 to 1997.

Measuring sustainabilitySustainability is defined, for the purpose of this chapter, as an improvement in theproductive performance of a system without depleting the natural resource base uponwhich future performance depends (Pandey and Hardaker 1995). Unsustainabilitymay result from on-site and/or off-site effects of agricultural land use. On-site effectsinclude adverse changes in the physical, chemical, and biological properties of thesoil-water-plant complex that reduce farm productivity. For example, in intensifiedirrigated rice systems of tropical Asia, reduced availability of nutrients to plants be-cause of changes in soil properties could lead to unsustainability (Cassman and Pingali1995). Off-site effects, which are also called externalities, refer to those effects thatare not normally valued in the market place. Common examples are adverse healtheffects of groundwater contamination and pesticide use, and damage to irrigationinfrastructure from soil erosion.

Two commonly used economic indicators of sustainability are trends in partialfactor and total factor productivities. Partial factor productivity is defined as the aver-age productivity of a factor of production. Total factor productivity (TFP) is definedas the ratio of the aggregate quantity of all outputs produced within a given timeperiod (usually a year) to the aggregate quantity of all inputs applied during the sametime period. Suitable weights based on prices or output and input shares are used foraggregating physical quantities of various outputs and inputs.

Economists consider TFP to be a more meaningful concept than partial factorproductivity for assessing sustainability (Lynam and Herdt 1989, Harrington 1993).As all inputs and outputs are accounted for, a declining trend in TFP is an indicator ofpossible degradation of the resource base, or unsustainability. Although the definitionrequires the inclusion of all inputs and outputs, data limitations and valuation prob-lems mean that only those inputs and outputs that can be easily measured and valuedare generally included. The externalities, such as environmental pollution, which aredifficult to value, are often excluded from TFP calculations. Similarly, changes inprices of inputs and outputs can affect TFP values over time, despite the use of meth-ods that attempt to correct for such price effects (Rayner and Welham 1995). Despitesome of these practical limitations to calculating TFP, the trend in TFP (not its level)is considered to be a useful indicator of (un)sustainability and has been widely used(Capalbo and Antle 1988).

Following the method suggested by Rayner and Welham (1995), we used theTornqvist-Theil method to calculate the TFP index. The Tornqvist-Theil method isconsidered to be theoretically superior to other methods since it is consistent with aflexible production function that does not arbitrarily constrain the substitution possi-bilities between inputs. The input index I(X)t is computed as

396 Lucas et al

where xit = quantity of input i in period t, sit = share of input i in total cost in period t;

wit, wkt = actual prices of inputs in period t. Similarly, the output index is computed as

where qjt = quantity of output j in period t, rjt = share of output j in total revenue inperiod t,

pjt, pit = actual prices of outputs in period t. Finally, the TFP index is obtained as theratio of I(Q)t / I(X)t.

The computation of these indices follows the following procedure. For eachyear, all crop outputs produced and all inputs used were included in the calculation.Analysis was limited to the period 1992-97 as the data for 1991 and 1998 were in-complete. A combination of province-level and farm-level data was used in the analy-sis as farm-level data for 1992-94 did not include all outputs for all farmers. On theoutput side, provincial-level data on yield and farm-gate prices for 1992-97 wereused. For the cost of production, data for individual crops for each year were notavailable at the provincial level. Hence, farm-level data were used for this purpose.Indices of the value of outputs and inputs were subsequently obtained using equa-tions (1) and (3).


Cropping patternsRice is grown in most of the farmlands during the WS, except for small areas in thehigher toposequence that are planted to vegetables such as beans, eggplant, and to-mato. Most of the farmers plant modern rice varieties. In the 1996 and 1997 WS,IR64 and PSB Rc14 were the most popular varieties planted, occupying 24% and25% of the rice area, respectively (Table 1). BPI Ri10 remained a common varietyamong farmers. Other IR varieties covered almost 25% of the total area planted in thelast two years.

During the DS, farmers plant high-value cash crops such as garlic, onion, to-mato, sweet pepper, and tobacco. These crops are entirely dependent on irrigationfrom shallow tube wells. Vegetables such as eggplant, bitter gourd, bottle gourd, squash,

sit = , i,k = 1, ... n, and (2)witxit

Σwktxkt

rjt = ; i,j = 1, ... m, and (4)pjtqjt

Σpitqit

I(X)t = I(X)t – 1 exp[1/2Σ (sit + si, t – 1) ln xit – ln xi, t – 1)] (1)i

I(Q)t = I(Q)t – 1 exp[1/2Σ (rjt + r j, t – 1) ln qjt – ln q j, t – 1)] (3)j


and cowpea are also planted in smaller parcels. Some farmers can even grow a thirdcrop.

Economics of rice productionRice is planted predominantly during the wet season. There is an increasing use ofmodern rice varieties. Traditional rice varieties are rarely found. Farming activitiesbegin immediately after rains occur. However, farmers usually experience unexpecteddry periods after a seemingly continued downpour. This forces them to resort to supple-mental irrigation. Land preparation is becoming more mechanized using hand trac-tors. This has decreased the labor requirement for land preparation by 25%. Waterbuffaloes, however, are still a common sight during land preparation.

Rice plants are established by transplanting, with the average seeding rate be-ing 101 kg ha–1 (Table 2). Farmers still use more than the recommended seeding rateof 40 kg ha–1. The average fertilizer rate was 129-31-21 kg NPK ha–1. Fertilizers wereapplied in two splits—2 wk after transplanting and 5 wk after transplanting. Basalapplication was seldom practiced. The rate of application of insecticides and herbi-cides was low. Labor use for crop management is generally low and accounts for onlyabout 14% of the total labor input.

The average grain yield of rice from 1996 to 1997 decreased by 16% comparedwith the mean yield for 1994 to 1995 (Table 3). The decrease in yield is attributed tothe El Niño phenomenon. The average returns above cash costs per hectare were$450. About 40% of the total production cost was accounted for by material inputs.Despite a decrease in total labor ha–1, labor cost increased because of a higher wagerate.

Economics of high-value cash cropsVarious dry-season cash crops are planted after rice. Over the years, maize occupiedalmost one-third of the total area in rainfed areas. Tobacco covered more area (7%)than tomato (5%) and sweet pepper (3%) from 1996 to 1998. This may be attributedto the increasing economic benefit from tobacco unlike in the past years. Area de-

Table 1. Rice varieties planted (1994-97), Ilocos Norte, Philippines.

Variety 1994 1995 1996 1997% area

IR64 18 26 24 15BPI Ri10 30 27 22 20PSB Rc14 – 2 13 26Other IR varietiesa 45 28 23 25Other PSB Rc varietiesb 2 11 15 10Other varietiesc 5 6 3 4

aOther IR varieties include IR36, IR42, IR58, IR60, IR66, IR68, and IR78. bOtherPSB Rc varieties are PSB Rc8, PSB Rc10, PSB Rc12, PSB Rc18, PSB Rc22,PSB Rc28, and PSB Rc34. cOther varieties include UPL Ri 4, UPL Ri 5, c4-137,C22, and glutinous.

398 Lucas et al

Table 2. Average material and labor inputs for rice (1994-97), Ilocos Norte, Philippines.

Categories 1994 1995 1996 1997 Allyears

Material inputsSeed (kg ha–1) 111 (76) 107 (68) 103 (53) 93 (42) 101 (57)Nitrogen (kg ha–1) 155 (87) 143 (99) 121 (41) 116 (39) 129 (64)Phosphorus (kg ha–1) 30 (25) 39 (38) 30 (22) 30 (19) 31 (25)Potassium (kg ha–1) 19 (15) 22 (20) 19 (18) 25 (19) 21 (18)Insecticide (kg ai ha–1) 0.03 (0.09) 0.05 (0.11) 0.09 (0.25) 0.03 (0.09) 0.05 (0.17)Herbicide (kg ai ha–1) 0.01 (0.08) 0.07 (0.17) 0.12 (0.22) – 0.05 (0.15)Fuel for land preparation 7 (28) 6 (14) 6 (21) 31 (29) 14 (27)

Labor inputs (person-days ha–1)Land preparation 8 (9) 8 (7) 7 (6) 6 (6) 7 (7)Crop establishment 28 (21) 41 (30) 35 (22) 26 (18) 32 (23)Crop managementa 6 (6) 12 (14) 8 (10) 18 (21) 11 (13)Harvesting and threshing 38 (32) 45 (29) 37 (20) 26 (19) 35 (25)Total labor 80 (49) 106 (57) 87 (58) 76 (43) 85 (47)

aIncludes fertilizer, chemical application, and weeding; ai = active ingredient. Numbers in parentheses arestandard deviations.

Table 3. Average grain yield, costs, and returns for rice (1994-97), Ilocos Norte, Philippines.

Categories 1994 1995 1996 1997 All years

Yield (t ha–1) 3.95 (2.4) 3.56 (1.4) 3.30 (1.3) 2.88 (1.2) 3.31 (1.5)

Material inputs (US$ ha–1)Seed 24 (16) 23 (15) 29 (15) 25 (12) 26 (14)Fertilizer 107 (58) 128 (72) 76 (27) 74 (26) 89 (48)Insecticide 6 (12) 7 (13) 3 (5) 3 (6) 4 (9)Herbicide 0.42 (2) 1.2 (3) 0.10 (1) 0.30 (1)Power/fuela 36 (47) 39 (36) 24 (30) 46 (27) 36 (35)

Labor inputs (US$ ha–1)Land preparation 24 (26) 24 (23) 20 (17) 44 (15) 33 (19)Crop establishment 59 (41) 82 (85) 94 (60) 94 (50) 103 (60)Crop managementb 16 (18) 33 (40) 28 (33) 5 (32) 10 (32)Harvesting and threshing 106 (89) 40 (87) 100 (53) 33 (52) 42 (70)Total material costs 174 (95) 198 (78) 131 (45) 147 (46) 155 (66)

(US$ ha–1)Total labor costs (US$ ha–1) 206 (122) 179 (123) 243 (110) 177 (98) 197 (114)Total costs (US$ ha–1) 380 (190) 377 (137) 352 (127) 324 (120) 351 (140)Gross returns (US$ ha–1) 998 (604) 899 (341) 569 (229) 549 (229) 688 (384)Returns above paid-out 855 (563) 799 (351) 277 (229) 249 (386) 450 (458)

costs (US$ ha–1)Net returns (US$ ha–1)c 618 (526) 522 (334) 194 (231) 225 (222) 336 (356)

aFor land preparation and irrigation. bIncludes fertilizer, chemical application, and weeding. cNet of cash cost andimputed cost of family labor; 1US$ = P25. Numbers in parentheses are standard deviations.


voted to sweet pepper remained the lowest. Fertilizer rates applied to dry-season cropsusually exceeded the recommended rates. The average fertilizer rate applied to sweetpepper was 305-85-78 kg NPK ha–1 and for garlic was 136-49-41 kg NPK ha–1 (Table4). The recommended rates are 170-57-163 kg NPK ha–1 for sweet pepper and 90-26-50 kg NPK ha–1 for garlic. However, from 1996 to 1998, about a 30% decline and10% decline in fertilizer application to sweet pepper and garlic were observed, re-spectively. These declines could be partly due to the increasing awareness of farmersregarding apparent pollution of groundwater. Some degree of consciousness may haveoccurred after information generated from several research studies in sweet pepper-growing areas started to spread (e.g., Gumtang et al 1999, Shrestha and Ladha 1998,Tripathi et al 1997). However, pesticides were applied at high rates. For example,sweet pepper is sprayed with pesticides weekly.

Material inputs account for 40% to 50% of the total production costs in dry-season crops (Table 5). The high marginal profitability of sweet pepper, tomato, andgarlic may have encouraged farmers to apply higher doses of inputs. Economic re-turns from dry-season crops showed that sweet pepper was the most profitable, fol-lowed by tomato and garlic with net returns of $1,096, $861, and $842 ha–1, respec-tively. High yields from sweet pepper and its higher price are factors that contributedto higher returns. Low returns from garlic are attributed to the dramatic drop in farm-gate prices (Fig. 6) due to the poor quality of output in 1997.

Table 4. Yield and input use of major dry-season crops (1994-98), Ilocos Norte, Philippines.

Categories Maize Garlic Mungbean Sweet pepper Tomato

No. of fields 184 233 135 54 89Yield (t ha–1) 3.15 (2.9) 0.84 (0.73) 0.41 (0.37) 6.0 (4.8) 33 (19.5)

Material inputsSeed (kg ha–1) 21 (15) 264 (119) 31 (20) 1.3 (1) 0.70 (0.59)Nitrogen (kg ha–1) 102 (73) 136 (71) 6 (30) 305 (138) 126 (51)Phosphorus (kg ha–1) 23 (25) 49 (35) 2 (9) 85 (69) 67 (42)Potassium (kg ha–1) 27 (30) 41 (40) 1.3 (8) 78 (69) 111 (73)Insecticide (kg ai ha–1) 0.09 (0.25) 0.21 (0.44) 0.18 (0.39) 1.5 (1.3) 0.58 (0.65)Fungicide (kg ai ha–1) 0.02 (0.15) 0.80 (1.6) 0.10 (0.37) 2.10 (3) 1.71 (2)Herbicide (kg ai ha–1) – 0.08 (0.44) – 0.07 (0.48) –Fuel (L ha–1)a 16 (20) 32 (32) 16 (19) 81 (52 ) 32 (23)

Labor inputs (person-days ha–1)Land preparation 3 (4) – 3 (6) 3 (6) 4 (7)Crop establishmentb 14 (19) 20 (20) 6 (5) 43 (42) 26 (20)Crop managementc 4 (4) 15 (13) 2 (3) 9 (9) 6 (5)Irrigation 4 (5) 6 (4) 3 (3) 13 (13) 4 (4)Weeding 1.3 (2) 10 (13) 1 (5) 14 (15) 2 (5)Harvesting and threshing 14 (16) 12 (14) 17 (22) 23 (30) 30 (30)Total labor 40 (35) 63 (39) 32 (37) 105 (65) 73 (50)

aFor land preparation and irrigation. bIncluding mulching in garlic. cIncludes fertilizer, chemical application, andweeding. Numbers in parentheses are standard deviations.

400 Lucas et al

Table 5. Costs and returns of major dry-season crops (1994-98), Ilocos Norte, Philippines.

Categories Maize Garlic Mungbean Sweet pepper Tomato

Material inputs (US$)Seed 36 (29) 607 (343) 32 (27) 109 (93) 93 (77)Fertilizer 67 (51) 134 (109) 5 (32) 239 (149) 155 (76)Insecticide 7 (16) 16 (32) 10 (19) 110 (115) 77 (64)Fungicide 2 (13) 12 (23) 2 (5) 43 (63) 54 (64)Herbicide 1 (7) 1 (1) 1 (5)Power/fuela 54 (38) 27 (15) 52 (64) 95 (33) 65 (25)

Labor inputs (US$)Land preparation 18 (24) 17 (28) 23 (31) 23 (34)Crop establishment 20 (31) 60 (59) 21 (41) 53 (37) 53 (49)Crop management 6 (8) 7 (7) 1 (3) 26 (31) 18 (14)Irrigation 14 (15) 23 (21) 11 (36) 73 (99) 21 (21)Weeding 7 (7) 61 (70) 2 (15) 130 (134) 12 (18)Harvesting and threshing 69 (64) 49 (126) 67 (67) 124 (102) 146 (411)Total material costs (US$) 156 (77) 786 (404) 80 (58) 547 (236) 383 (181)Total labor costs (US$) 133 (95) 200 (156) 118 (105) 429 (211) 272 (411)Total costs (US$) 289 (124) 986 (450) 198 (134) 976 (343) 655 (504)Gross returns (US$) 654 (608) 1,828 (2,021) 310 (285) 2,057 (2,047) 1,517 (941)Net returns (US$)b 366 (577) 842 (1,962) 112 (294) 1,096 (2,134) 861 (936)

aFor land preparation and irrigation. bNet of cash cost and imputed cost of family labor. 1US$ = P25. Numbersin parentheses are standard deviations.

Fig. 6. Index of real farm-gate prices of major crops, IlocosNorte, Philippines, 1991-98. Source: Bureau of AgriculturalStatistics (1999).

250

200

150

100

50

01991 1992 1993 1994 1995 1996 1997 1998

Year

Price index

MaizeRiceMungbeanSweet pepperTomatoGarlic


Trend in total factor productivityThe index of input use generally increased but seemed to have leveled off in 1996 and1997 (Fig. 7). The output index, which initially showed an increasing trend, declinedafter 1995. The decline in output index can be partly attributed to the prolonged droughtand strong typhoons experienced in the province in the past three years. The declinein output may also be due partly to the decline in the yield of high-value crops such astomato and garlic. The time period covered in the analysis is too short to detect anylong-term trend in total factor productivity. The fluctuations observed in total factorproductivity likely reflect mainly the climatic variation.

Contamination of groundwater with NO3-N in intensively cultivated areas wasobserved. Results of groundwater monitoring showed that, in areas where rice-sweetpepper is practiced, NO3-N concentration was consistently high in both the WS andDS (Alibuyog et al 1999). A high NO3-N concentration was also found in areas plantedto garlic, tobacco, and tomato. In addition to NO3-N contamination of groundwater,there were some indications of salinity intrusion because of excessive pumping. Elec-trical conductivity (EC) was observed to be generally high. Although a decline wasobserved at the start of the WS, the EC again increased toward the end of the DS. Theincrease in EC could be due to possible intrusion of saline water in the area as anexcessive quantity of irrigation water was used during the DS (Alibuyog et al 1999).In sweet pepper, irrigation interval is one week with a total irrigation depth of 397mm per season. Garlic is irrigated 4 to 5 times per season with a total irrigation depthof 235 mm, while tobacco, tomato, maize, and eggplant are irrigated 2 to 3 times perseason with depths of 129, 214, 187, and 92 mm, respectively (Alibuyog et al 1999).

Conclusions

The trend in total factor productivity was positive (1992-95) initially but becamenegative afterwards (1996-97). Indications of contamination of groundwater with NO3-N and excessive drawdown of groundwater were also apparent. If these negative ex-ternalities continue to increase, production systems in Ilocos Norte may not be sus-

Fig. 7. Trend in total factor productivity (TFP), in-put indices, and output indices for Ilocos Norte,Philippines, 1992-97.

200

160

120

80

40

01992 1993 1994 1995 1996 1997

TFP, output and input indices

Input indexOutput indexTFP

402 Lucas et al

tainable in the future even though total factor productivity did not show any clearnegative trend over the study period.

Without appropriate interventions, the intensive cropping systems in Ilocos Nortemay become unsustainable in the long run. A proper understanding of N and irriga-tion requirements of crops as well as timing of application is necessary to avoid ex-cessive losses of N and to maintain sustainability and environmental quality (Bucaoet al 1999). Farmers should be encouraged to consider other crops such as N catchcrops (e.g., maize + indigo, mungbean) as alternatives to continuous rice-cash crop.The maize + indigo intercrop during the dry-to-wet transition has been effective incapturing residual soil N (Alam 1999). Extensive dissemination of information on thenegative effects of high NO3-N concentration in the groundwater should be carriedout. Research and extension programs that improve the quality of farm managementpractices so that nutrient-use efficiency is increased and input costs are reduced areneeded to ensure the sustainability of the production system.

ReferencesAlam M. 1999. Increasing yield and nutrient use efficiency through improved fertilization and

integrating of catch crop in a rice-vegetable cropping system. Unpublished PhD thesis,University of the Philippines Los Baños.

Alibuyog NR, Bucao DS, Agustin EO, Tuong TP. 1999. Annual report. Rainfed Lowland RiceResearch Consortium. Batac, Ilocos Norte (Philippines): Mariano Marcos State Univer-sity.

Bucao DS, Gumtang RJ, Alibuyog NR, Agustin EO, Tuong TP, Obien SR. 1999. Crop diversi-fication and intensification impact on groundwater resource. Paper presented during theCommodity Review held at CONDORA, Damortis, La Union, Philippines, 27-28 May1999.

Bureau of Agricultural Statistics. 1998. Production survey of rice and major dry season cropsin the province of Ilocos Norte. Bureau of Agricultural Statistics, Laoag City, IlocosNorte, Philippines.

Bureau of Agricultural Statistics. 1999. Monthly average farm gate prices of selected agricul-tural commodities, Province of Ilocos Norte, 1991-1998. Laoag City, Ilocos Norte, Phil-ippines.

Capalbo S, Antle JM. 1988. Agricultural productivity measurement and explanation. Washing-ton, D.C. (USA): Resources for the Future.

Cassman KG, Pingali PL. 1995. Intensification of irrigated rice systems: learning from the pastto meet future challenges. GeoJournal 35:299-305.

Gumtang RJ, Pampolino MF, Tuong TP, Bucao D. 1999. Groundwater dynamics and qualityunder intensive cropping systems. Exp. Agric. 35:153-166.

Harrington LW. 1993. Interpreting and measuring sustainability: issues and options. In:Harrington LW, Hobbs PR, Cassaday KA, editors. Methods of measuring sustainabilitythrough farmer monitoring: application to the rice-wheat cropping pattern in South Asia.Proceedings of the Workshop. Mexico: CIMMYT, IRRI and NARC.

Lucas M, Pandey S, Villano R, Culannay D, Obien SR. 1999. Characterization and economicanalysis of intensive cropping systems in rainfed lowlands of Ilocos Norte, Philippines.Exp. Agric. 35:211-224.


Lynam JK, Herdt RW. 1989. Sense and sustainability: sustainability as an objective in interna-tional agricultural research. Agric. Econ. 3:381-398.

Pandey S, Hardaker JB. 1995. The role of modelling in the quest for sustainable farming sys-tems. Agric. Syst. 47:439-450.

PPDO (Provincial Planning Development Office). 1995. Comprehensive land use plan. Prov-ince of Ilocos Norte (Philippines): PPDO.

Rayner AI, Welham SJ. 1995. Economics and statistical considerations in the measurement oftotal factor productivity (TFP). In: Barnett V, Payne R, Steiner R, editors. Agriculturalsustainability: economic, environmental and statistical considerations. New York (USA):John Wiley. p 23-28.

Shrestha RK, Ladha JK. 1998. Nitrate in groundwater and integration of a nitrogen catch cropin intensive rice-based cropping systems to reduce nitrate leaching. Soil Sci. Soc. Am. J.62:1610-1619.

Tripathi BP, Ladha JK, Pandey S. 1997. Economic feasibility, production potential and nitro-gen behavior in intensively cultivated rice-based cropping systems in Northern Luzon,Philippines. Philipp. J. Crop Sci. 22:39-48.

NotesAuthors’ addresses: M.P. Lucas, D.R. Culannay, T.F. Marcos, Mariano Marcos State Univer-

sity, Batac 2906, Ilocos Norte, Philippines; S. Pandey, R.A. Villano, Social SciencesDivision, International Rice Research Institute, DAPO Box 7777, Metro Manila, Philip-pines.


Integrating biophysicaland socioeconomic

characterization

406 Garcia et al

Socioeconomic and biophysical characterization of rainfed . . . 407

Myanmar’s rainfed rice land covers 79% of its total cultivated rice area, whichtranslates into about 4 million hectares, one of the largest in the world.Because of a lack of irrigation facilities, rice is generally grown once a yearduring the monsoon season. A study in the Ayeyarwardy Delta, a major rice-producing area in southern Myanmar, was conducted to characterize the dif-ferent production systems of rice farmers. The study also investigated theoperations of the output and input (i.e., land, labor, and capital) markets andhow farmers’ access to these markets shaped the distribution of income andresources in the village economy.

The biophysical and socioeconomic factors leading to the adoption ofnew rice-based technologies, such as summer rice, double monsoon rice,the use of high-yielding varieties, and rice-fish culture, were identified. Fur-thermore, biotic and abiotic factors that significantly constrained rainfed riceproduction were evaluated in terms of reduction in rice yield and productivity.Alternative management practices and policy options were proposed to helpminimize the adverse effects of these constraints.

Myanmar, one of the last frontiers for increasing world rice production, is located inSoutheast Asia. Thailand, Lao PDR, China, India, and Bangladesh surround its bor-ders from east to west. Its land area of 68 million hectares spans 2,092 kilometers inlength from north to south and about 925 kilometers from east to west at its widestparts. In 1997, the total population was 46 million, with 75% living in the rural areas.

The country’s economy basically depends on agriculture, with about 15% of itsarea currently used for crop production. Unlike many countries in Asia, it has a largepotential for increasing the cropping area by opening untapped culturable wastelands,which are approximately 12% of its total land area. Rice, the main staple food crop ofthe country, is grown extensively, covering 53% of the country’s total area sown tocrops. Rice production accounts for 34% of the gross domestic product and 47% ofthe total agricultural exports. It is also one of the principal sources of foreign ex-change, contributing 57% of the total export earnings.

Socioeconomic and biophysicalcharacterization of rainfedversus irrigated rice productionin MyanmarY.T. Garcia, M. Hossain, and A.G. Garcia

408 Garcia et al

Increased rice production had always been the government’s major thrust incrop production. Rice policies were geared toward meeting not only the local ricedemand but also providing surplus production for export. In the early 1990s, the gov-ernment launched a nationwide campaign to boost rice production. To achieve theseobjectives, the Ministry of Agriculture and Irrigation focused on the following strat-egies: (1) expand cultivated areas by opening new frontiers, (2) explore potentialwater resources for irrigation facilities, (3) intensify the use of modern rice varieties,and (4) adopt new cropping practices to enhance production, such as summer rice,double monsoon rice, and rice-fish culture.

The success of these strategies depends very much on how farmers respond tothe government’s policies. Farmer responses, however, depend to a large extent onthe biophysical and socioeconomic factors that influence farmers’ adoption of thenew rice-based technologies. Rice production can be increased if major biotic andabiotic constraints can be properly identified and appropriate technologies to solvethem can be recommended.

The main objectives of this chapter are (1) to document the new farming prac-tices that evolved out of recent developments in the rice sector and (2) to identify thebiophysical and socioeconomic opportunities for and constraints to the adoption ofthe new rice-based technologies, such as summer rice, double monsoon rice, the useof high-yielding varieties, and rice-fish culture. The chapter also investigates the op-erations of the output and input (i.e., land, labor, and capital) markets and their effectson the distribution of income and resources in the village economy. Based on thefindings, the chapter proposes alternative management practices and policy optionsthat could help alleviate these constraints.

Methodology, data sources, and surveys

Most of the data at the national level used in this study were from the MyanmarAgriculture Service (MAS) of the Ministry of Agriculture and Irrigation. A surveywas also conducted in 1996 to study the rural household economy and documentfarmers’ responses to the government’s rice production programs. The study coveredfour selected villages in Nyaungdong Township located in the Ayeyarwardy Delta.The Ayeyarwardy Delta, located in southern Myanmar, is the largest rice-producingregion and contributes about 34% of the country’s total rice production. Approxi-mately 1.8 million hectares of its land are allocated to rice production with an averageyield of 3.2 t ha–1.

The villages were chosen according to the presence of new rice-based croppingsystems practiced by farmers in the area, namely, (1) double rice cropping (monsoonrice followed by summer rice), (2) triple rice cropping (two rice crops in the monsoonseason and one summer rice crop), (3) rice-fish farming, and (4) one rainfed ricecropping during the monsoon season. The sample was generated through total enu-meration consisting of 739 households, of which 40% were landless. Only 22% of thetotal farm households were observed to adopt the new rice-based technologies pro-moted by the government, that is, double cropping with summer rice as a second crop


(13%), triple rice cropping (3%), and rice-fish farming (6%). Hence, the remaining78% of the total farming households were all rainfed farms, which grew only one ricecrop during the monsoon season. Despite the low percentage of adopters of the newlyestablished rice-based technologies, their experiences provided the study with valu-able insights regarding the incentives, constraints, and problems they faced in theprocess of adoption and adaptation of these technologies.

Farmer interviews were carried out to identify biophysical and socioeconomicfactors that led to the adoption of the new rice-based technologies. A statistical ap-proach and cost and return analysis were used to characterize the impacts of recentdevelopments on the village rice economy and farmers’ production systems. Like-wise, Gini ratios were estimated and Lorenz curves were fitted to describe the incomedistributions found in the sample villages representing the different rice ecosystemsin the study area.

To identify the major rice production constraints, data were collected through aseries of surveys in 30 selected townships from five divisions and one state(Ayeyarwardy, Bago, Yangon, Sagaing, and Mandalay divisions, and Shan State) inupper, central, and lower Myanmar. These sites were selected since approximately80% of the total rice crop in Myanmar is grown in these locations. Major biotic andabiotic factors that significantly constrained rice production were evaluated and rankedin terms of yield reduction brought about by these factors. Problems associated withthese constraints were then prioritized, which allowed us to identify alternative man-agement practices that can help minimize their adverse effects.

Recent developments in the Myanmar rice economy

Sown areaThe crop area planted to rice was less than 5 million hectares from 1962 to 1990(Table 1). Since 1991, however, area planted to rice grew annually by 4%, reaching 6million hectares in 1995. Government efforts to expand and open new lands increasedthe net sown area by one million hectares over five years. In the crop year 1995-96,total rice area for the whole Union of Myanmar was 6 million hectares, wherein 4.83million hectares were devoted to monsoon (wet-season) rice and 1.21 million hect-ares for summer (dry-season) rice. After 1996, summer rice area started to decline,reaching only 0.89 million hectares in 1998, thus decreasing the total rice sown area.The summer rice area decreased because of the limited supply of diesel fuel for irri-gation pumps and commercial fertilizer (both related to foreign currency limitations),hence only the more suitable rice areas were targeted for planting.

Production and yieldRice production changed drastically from 1962 to 1995 (Table1). It increased fromabout 8 million metric tons in the 1960s to about 16 million metric tons in the early1990s. Average rice yields also almost doubled from 1.6 to 3.0 t ha–1 within the pe-riod. The production growth rate was about 1% in the 1960s and reached 5% in the1970s. The increase in rice yield was the major factor that induced production growth

410 Garcia et al

Table 1. Total rice area, average yield, total production, and fertilizer consumption from 1962to 1998 for the Union of Myanmar.

Total area Av yield Production Fertilizer UreaNo. Year (million ha) (t ha–1) (million t) consumption consumption

(t) (t)

1 1962-63 4.64 1.64 7.63 8,800 5,456 2 1963-64 4.86 1.59 7.75 8,841 5,865 3 1964-65 4.96 1.71 8.47 13,727 7,647 4 1965-66 4.83 1.66 8.02 6,962 4,187 5 1966-67 4.50 1.47 6.61 8,615 4,537 6 1967-68 4.69 1.65 7.73 12,245 7,971 7 1968-69 4.75 1.68 7.99 40,286 28,607 8 1969-70 4.65 1.71 7.95 26,179 18,926 9 1970-71 4.79 1.70 8.12 21,781 16,59010 1971-72 4.75 1.71 8.14 47,197 29,78211 1972-73 4.51 1.62 7.32 67,223 43,96512 1973-74 4.86 1.76 8.56 68,905 47,20213 1974-75 4.87 1.76 8.54 73,775 60,16714 1975-76 5.01 1.83 9.16 86,959 65,25515 1976-77 4.89 1.90 9.27 89,629 76,26816 1977-78 4.85 1.94 9.42 105,891 85,65917 1978-79 4.99 2.10 10.48 160,519 131,40718 1979-80 4.43 2.35 10.40 173,531 129,84019 1980-81 4.78 2.77 13.25 205,315 147,09820 1981-82 4.79 2.94 14.08 225,732 169,13421 1982-83 4.55 3.15 14.30 281,831 198,44022 1983-84 4.64 3.06 14.22 331,969 228,84223 1984-85 4.58 3.09 14.19 304,571 211,39624 1985-86 4.64 3.07 14.25 324,972 229,04125 1986-87 4.65 3.02 14.06 304,317 208,70426 1987-88 4.47 3.04 13.57 202,815 162,39027 1988-89 4.51 2.90 13.10 153,565 123,37328 1989-90 4.72 2.91 13.74 131,888 102,30729 1990-91 4.74 2.93 13.90 109,098 86,64730 1991-92 4.56 2.88 13.14 99,802 77,02531 1992-93 5.04 2.93 14.77 149,745 122,685

Monsoon rice 4.71 2.93 13.83Summer rice 0.32 2.87 0.93

32 1993-94 5.47 3.05 16.68 248,423 194,672Monsoon rice 4.65 2.97 13.81Summer rice 0.82 3.51 2.87





Source: Myanma Agriculture Service (1997).


during this period, brought about by the widespread adoption of modern rice varietiesand increased fertilizer use. In the 1980s, however, the growth rate plunged to lessthan 1%. This production decline was mainly due to (1) scarcity of fertilizer inputs,which was attributed to the weakening of the country’s economy since 1985, and (2)restricted access to foreign exchange, which severely hampered the importation ofagricultural imports. The early 1990s, on the other hand, saw dramatic changes inproduction growth rates, which soared to an average of 12% annually. This was broughtabout by the government’s program on summer rice production, which started in 1992,resulting in a significant increase in the total rice output of the country.

Farming practices adopted by rice farmers

Rainfed lowland rice-cropping systemDuring the crop year 1997-98, the total area planted to rainfed rice was approxi-mately 4.6 million hectares, which was estimated to be 79% of the total cropped areaof the country. It was generally grown under several conditions: normal plains (68%),deepwater area (7%), and saline area (4%).

Under the rainfed lowland ecosystem, only one rice crop was grown annuallysince production was totally dependent on rainfall as the main water source. Landpreparation normally began during the onset of the rainy season, mostly from May toJune when enough water had soaked the field. During field preparation, rice seedswere sown in seedbeds to grow seedlings. A majority of the rainfed lowland farmersused draft animals (a pair of bullocks) as a power source for land preparation andspent about 36 man-animal-days to finish one hectare.

Transplanting was the most common method of crop establishment used byfarmers, normally done in rows with the use of strings as guides. Farmers transplantedrice seedlings that were about 30 to 45 d old between June and July. Older rice seed-lings that were taller were generally preferred by farmers, especially those who didnot have water control in their fields. Taller seedlings have a higher survival rateduring heavy rains when fields become totally submerged in water.

Rice harvesting started in October to December. Farmers normally left the cutstraws to dry in the field before threshing. Harvesting done during the late monsoonwhen rainfall was still high posed a big problem in grain drying. After the rice cropwas harvested, the fields were left idle and became grazing grounds for livestockduring the dry season. Threshing, on the other hand, was done mostly with animals orby renting a mechanical thresher and paying a fixed amount per unit volume of ricethreshed.

Summer riceRice planted between September and February and harvested between December andApril was classified as the summer rice crop. As with rainfed rice, the most commonmethod of crop establishment was transplanting. More and more farmers, however,were learning to use direct seeding (done through either broadcasting or line seeding)to reduce the time constraint brought about by double cropping and to maximize the

412 Garcia et al

use of water by avoiding seedbed preparation for rice seedlings. Generally, farmerspracticing summer rice cultivation planted high-yielding varieties and applied higherrates of fertilizers; hence, summer rice yields were often higher than those of themonsoon rice (3.6 vs 3.0 t ha–1) if the water supply was adequate.

Large-scale promotion of summer rice cropping, which started in 1992, wasconsidered as one of the biggest achievements of the government in boosting riceproduction in the country. With an initial area of 0.32 million hectares and total pro-duction of 0.93 million metric tons, summer rice production had peaked at 4.0 milliontons in 1995, covering a total area of 1.2 million hectares (Table 1). In 1997, totalsown area declined to 0.89 million hectares, thereby reducing total production to 3.2million tons. The current target area for summer rice production is 1.6 million hect-ares with a total production of 5.0 million tons. To achieve this target, present irriga-tion facilities were being expanded since summer rice can only be planted whereirrigation water is available. As an incentive for farmers to grow summer rice, thegovernment waived the production quota (12 baskets acre–1 or 593 kg ha–1) that theyneeded to sell to the government’s cooperative for this season.

Two rice crops during the monsoon seasonAnother innovation for increasing rice production in Myanmar was the introductionof two rice crops during the wet season, which was implemented in 1992. Rice grownfrom May to September was considered the first monsoon crop, whereas rice grownfrom September to January was considered the second monsoon crop. This produc-tion system allowed farmers to achieve three rice crops per year, that is, one summerrice and two monsoon rice crops. Growing two monsoon rice crops, however, wasonly possible under good water control and drainage in the field, especially duringthe height of the monsoon rains when flooding normally occurred.

The common method of crop establishment in the first monsoon crop was trans-planting rice seedlings in rows. For the second monsoon rice crop, planting was nor-mally done immediately after harvesting the first rice crop. Because of time con-straints, farmers either practiced direct seeding or purchased rice seedlings for trans-planting since they had no time to grow their own seedlings.

Harvesting the first monsoon rice crop was normally done between Septemberand October, which often coincided with heavy rainfall such that farmers commonlyexperienced problems in threshing and grain drying. These postproduction problemsresulted in low yield and poor grain quality. Higher grain yields, on the other hand,were normally reported during the second monsoon rice crop.

Labor shortage, a common problem of rice farmers in Myanmar, was furtheraggravated under this technology since harvesting and other postproduction activitiesin the first monsoon crop coincided with land preparation for the second monsooncrop. Peak harvesting for the second rice crop was done in early December, whichalso overlapped with land preparation for the summer rice crop. Because of time andlabor constraints, few farmers adopted triple cropping of rice.


Rice-fish farmingA new technology introduced by the Ministry of Agriculture in the early 1990s wasthe rice-fish technology. It was implemented in the low-lying areas of the AyeyarwardyDelta with an allocated area of 1,600 ha. Each rice-fish pond consisted of a 2-ha plotwith 1.2 ha in the middle used for paddy production, 0.4 ha surrounding the paddyfield as pond area to rear fish, and 0.4 ha for embankment. Catfish, common carp, andprawns were usually grown in the pond. Vegetables, flowers, banana, and perennialfruit trees (guava, papaya, citrus, drumstick, etc.) were planted in the embankment.The rice-fish technology was envisioned to enable double cropping of rice with waterfrom the pond serving as irrigation during the dry summer months.

The rice-fish ponds were constructed by the government and sold to interestedparties who wished to practice the rice-fish technology, such as landless householdswith financial resources, retired government personnel, and key officials in the localgovernment. Each pond was sold at US$500 in 1992. In most instances, the buyers ofthe ponds obtained a medium-term loan from the government banks payable in threeyears with an interest rate of about 1.5% per month or 18% per annum.

The government targeted 120 ponds to be built in the study area for a period of5 years. The site was called Kan Taya (hundred ponds), where 112 ponds were actu-ally built. The number of ponds owned by operators ranged from 1 to 11 normallylocated side by side. The project became very popular in the pilot areas such thatprivate ponds were voluntarily constructed and operated by local farmers. It was esti-mated that about 5,300 ha or 33% of the total rice-fish ponds in the country wereconstructed under this system.

The rice-fish area increased rapidly from 1991, reaching maximum adoption in1993. In 1995, the government phased out the construction of rice-fish ponds. Sev-eral reasons were cited for the phase-out of the project: the enormous cost (both im-plicit and explicit costs) of pond construction and the implementation of a new rulethat allowed multinational companies to develop large tracts of land (as much as 400ha) for rice production. As an incentive for these companies, 50% of their rice pro-duce can be exported privately to generate export earnings needed to finance theirimportation of raw materials and agricultural inputs.

The existing ponds are still operational but plagued with problems. During themonsoon season, heavy rain normally flooded the rice-fish area, which equalized thewater level inside and outside the pond embankment. This made monsoon rice pro-duction and seeding of the pond with fingerlings impossible. Only when the waterlevel receded could rice production be started, often late in the monsoon season. Thebigger problem of the rice-fish technology lay in the fish culture. At the height of themonsoon rains, the seeded fingerlings escaped with the floodwater, which signifi-cantly reduced the fish population in the pond. This brought tremendous losses to theoperators.

Moreover, the cost of feed meal for the fish was often unaffordable for thesmall operators. Hence, the original technology of rice-fish culture was altered by thepond operators. Every monsoon season, a majority of the pond owners opened theirembankments and waited for the nearby rivers to overflow in their ponds. The flood-

414 Garcia et al

water often brought with it a variety of fish population that grew with minimum careand feeding. When the floodwater receded (in November to December), water fromthe ponds was pumped out to irrigate the rice fields in the middle of the plot forsummer rice cultivation. This allowed the harvesting of the wild fish trapped in theponds. Small operators find this practice viable since they can earn as much as $1,600from the sale of the wild fish harvest, on top of their earnings from summer riceproduction.

On the other hand, big-time operators of the rice-fish ponds altered the technol-ogy differently. They totally stopped rice production and concentrated on fish culture.Some actually dug up the middle plot for a larger pond area. Since the owners wereoriginally businessmen and retired government officials, they found rice growing tootiresome and laborious. Instead, they simply seeded their ponds with fingerlings anddiligently fed their fish stock until they were ready for harvest, which normally tookabout 1 1/2 to 2 years. Earnings from this practice ranged from $2,500 to $3,000 perpond, which were higher than the earnings from rice-fish culture per year with lesslabor involved.

Biophysical and socioeconomic characteristics of the village rice economy

Climate and rainfallMyanmar has two distinct seasons: (1) the dry season, which lasts from mid-Octoberto mid-May, and (2) the wet season from mid-May to mid-October. During the dryseason, there is a cold spell from December to February, after which the warm weatherbegins. Humidity is normally high from April to December. In the Ayeyarwardy Delta,the temperature is relatively less variable during the different seasons, ranging from16 to 37 °C compared with the northern parts of the country.

Agriculture depends highly on the southwest monsoon, which often occurs frommid-May to mid-October. During this period, rainfall provides sufficient moisture forgrowing crops except in the dry regions of northern Myanmar. This period also servesas the sole source of water to replenish dams and reservoirs. The amount of precipita-tion, however, varies significantly depending on the location. Higher annual rainfallis recorded in the coastal and deltaic regions, whereas low annual rainfall is observedin the dry and mountain regions.

Topography and irrigationVigorous public investments in irrigation infrastructure, such as the construction ofsmall to large dams and installation of pumping stations along the banks of majorwaterways, increased the area under irrigation in all of Myanmar from 12% of thetotal cultivable land in 1992 to 18% in 1995 (Table 2). This significantly increasedcropping intensity in the areas covered by irrigation development especially in thedelta area. About 80% of the country’s total irrigated area was devoted to rice produc-tion. Despite the government efforts, however, about 79% of the total rice area wasstill planted to rainfed rice during the monsoon season when rainfall was the onlywater source.


Topography in the study area was characterized by gentle rolling plains, whichcan be divided into three elevation zones: upper terrace, lower terrace, and deepwaterareas. These topographical zones were used to classify the irrigated villages at thestudy site. About 25% to 36% of the total sample farms were covered by irrigation(Table 3). Water was generally drawn using lift pumps from irrigation canals, creeks,rivers, and streams. Some of the irrigated farms (12% to 31%) drew water by pump-ing from irrigation canals that were constructed manually by the community of farm-ers in the villages under the leadership of the village chairman.

Ownership and characteristics of land resourcesTable 4 presents the distribution of landholdings in the sample villages. The averagelandholding size ranged from 1.8 to 2.5 ha but many farming households (35% to49%) owned less than 2 ha of land. On the other hand, the percentage of landlessranged from 19% to 51% of the total households. The lower percentage of landless inthe deepwater villages was attributed to the existence of common lands that werecharacterized by open access. These marginal lands were located in the deepest por-

Table 2. Total irrigated area from 1961 to 1995,Union of Myanmar.

Net sown area IrrigatedYear (000 ha) area Percentage

(000 ha)

1961-62 7,136 534 81971-72 7,933 887 111981-82 8,383 1,040 121991-92 8,308 995 121992-93 8,683 1,106 131993-94 8,706 1,332 151994-95 8,962 1,639 18

Source: MAI (1995).

Table 3. Distribution of land area by source of irrigation, Nyaungdon Township, 1996.

Irrigated

Rainfed Lower terrace Upper terrace DeepwaterSource of irrigation

Total area % Total area % Total area % Total area %(ha) area (ha) area (ha) area (ha) area

Not irrigated 145 99 182 64 112 75 276 69Pump from canals – – 91 31 23 16 48 12Tube wells – – – – – – – –Creeks/streams/ 1 1 13 5 14 9 76 19

riversTotal 146 100 286 100 149 100 400 100

416 Garcia et al

tion of the village and were generally submerged throughout the monsoon season.During the dry season, however, residual moisture in the soil allowed cropping ofrice, groundnuts, and pulses.

About 93–98% of the farmers reported that they owned the land they farmedeither legally or de facto as in the case of common lands (Table 5). Absolute landownership in the country was vested in the state. Farmers, however, were granted theright to cultivate the land and reap its benefits. This right was supported by a certifi-cate that could be transferred from generation to generation. Since land ownershipwas not absolute, transferability through buying and selling was restricted. This ren-dered the market for agricultural lands nonfunctional. However, illicit buying andselling of land were reported to exist in the villages at the rate of $60 ha–1. A very

Table 4. Distribution of farm households by size of landholding,Nyaungdon Township, 1996.

Irrigated

Landholding (ha) Rainfed Lower Upper Deepwaterterrace terrace

(% of HHa)

No cultivated land 48 44 51 190.01–1.0 16 13 15 241.01–2.0 21 22 22 252.01–3.0 6 4 6 103.01–4.0 6 8 5 94.01–10.0 3 8 1 1110.0 and above 0 1 0 2Total 100 100 100 100Average size of 1.77 2.53 1.71 2.48

holding (ha)

aHH = households.

Table 5. Distribution of farm households by tenure status, NyaungdonTownship, 1996.

Irrigated

Tenure status Rainfed Lower Upper Deepwaterterrace terrace

(% of HHa)

Owned farms 98 93 99 95Tenant – 2 – 1Leaseholder 2 3 1 3Rented out – 2 – 1Total 100 100 100 100

aHH = households.


small percentage of the land (1–3%) was cultivated under a leasehold arrangement,mostly between families who lent and borrowed parcels of land for a given season.Land leasing and tenancy arrangements were not regularly practiced in the study areaand elsewhere.

During the height of the monsoon season, low-lying areas, that is, deepwaterand lower terrace areas, were usually submerged. Floodwater normally rose to atleast 90 cm, covering an area of 69% and 66%, respectively (Table 6). In the rainfedvillage, however, flooding was not a big problem. Only 35% of the land area wasaffected by flood, of which 25% fell under 30 cm of water. Flooding also occurred inthe upper terrace areas, but with an even distribution of floodwater.

Farmers’ assessments of soil quality conditions in their fields were solicited.Categories such as good, average, poor, and very poor were given as choices. Most ofthe farmers (82%) in the rainfed village claimed that soil quality on their farms wasgood (Table 7). In contrast, many of the farmers owning irrigated farms characterizedtheir lands as having poor soil quality (52–66%). This was expected since extensivecropping of the field can easily exhaust soil fertility. On average, about 14% of thehouseholds in all four villages considered their land to be of average quality. Cases ofvery poor soil quality were reported in the deepwater area, but were of negligibleproportion.

Table 6. Distribution of land area (in ha) by depth of flooding, Nyaungdon Township, 1996.

Irrigated

Depth of Rainfed Lower terrace Upper terrace Deepwaterflooding


Not flooded 94 65 21 7 57 38 41 10Up to 30 cm 37 25 8 3 26 17 12 330–90 cm 5 3 68 24 31 21 74 18>90 cm 11 7 190 66 36 24 274 69

Table 7. Distribution of land area (in ha) by soil quality, Nyaungdon Township, 1996.

Irrigated

Rainfed Lower terrace Upper terrace DeepwaterSoil quality


Good 120 82 51 18 53 35 97 24Average 17 11 47 16 19 13 66 16Poor 8 6 188 66 78 52 236 59Very poor 1 1 – – – – 2 1

418 Garcia et al

Soil types found in the villages were classified into sandy loam, clay loam,clay, and silty. Many of the households (50–59%) claimed that clay loam was thepredominant soil type in the four villages (Table 8), followed by sandy loam (25–35%). There were reports of clayey soil (15–23%) but mostly in the irrigated areas.Clayey soil was not conducive to rice production since the soil was deemed too hardfor land preparation. This could have accounted for the farmers’ perception that soilquality was generally poor in the irrigated areas. Incidence of silty soil was reportedonly in the deepwater area, but was likewise negligible in proportion.

Rice varieties (traditional vs modern)The adoption of high-yielding modern varieties was another important strategy thatthe government actively pursued in conjunction with the introduction of intensiverice cultivation. Table 9 presents the distribution of farmers using modern and tradi-tional varieties in the survey area during the monsoon and summer seasons. Amongfarmers who engaged in rice double cropping, that is, rice-fish and summer rice, theuse of modern varieties was normally higher in the dry season (95%) than in the wetseason (34%). The wet-season crop was normally planted with traditional varieties(66%). This was generally attributed to the geographical location of most summerrice fields, where excessive flooding occurred during the height of the monsoon sea-son, hence the higher adoption of the traditional tall varieties. The traditional variet-

Table 8. Distribution of land area (in ha) by soil type, Nyaungdon Township, 1996.

Irrigated

Soil type Rainfed Lower terrace Upper terrace Deepwater


Sandy loam 50 35 72 25 42 28 37 34Clay loam 86 59 149 52 76 51 199 50Clay 9 6 65 23 32 21 62 15Silty – – – – – – 2 1

Table 9. Percent adoption of modern versus traditional varieties byfarming practices and season, Nyaungdon Township, Myanmar, 1996.

Rice- Two Irrigatedfish monsoon (double

Variety Rainfed farming rice cropping) Average

WSa WS DS WS1 WS2 WS DS WS DS

Traditional 22 65 2 15 0 67 8 34 5Modern 78 35 98 85 100 33 92 66 95

aWS = wet season, DS = dry season.


ies also produced more rice straws, which were often sold or used as livestock feed.The preference for traditional varieties was further reinforced by their desirable quali-ties: better eating quality, higher rice volume when cooked, and higher price in themarket. Harvests from traditional varieties were normally kept for home consump-tion or as a ready source of cash.

On the other hand, farmers who practiced triple rice cropping used more mod-ern varieties in all three seasons because of the short turn-around time needed tomake three croppings possible. Hence, shorter-duration varieties (a characteristic ofmodern varieties) were more preferred. Likewise, rainfed farmers planted more mod-ern varieties (78%) than traditional varieties to maximize yield for the only seasonthey grew rice. Farmers used a wide range of varieties in the study area for bothtraditional and HYVs (16 and 19 varieties, respectively).

Farm mechanizationBecause of the rapid expansion in cultivated area and increased potential for doublecropping brought about by irrigation infrastructure development, farm mechaniza-tion became an integral part of the government’s strategy for boosting agriculturalproduction. Labor shortage was one of the biggest problems of farmers, especiallyduring planting and harvest, when labor demand peaked. To solve this problem, theAgricultural Mechanization Department (AMD) put up 25 tractor stations in selectedtownships all over the country where needed machines, such as tractors, power tillers,high-lift pumps, threshers, seeders, and harvesters, were supplied to farmers at a fixedrent.

The government also encouraged private hiring of farm machinery. To boostlocal participation in the marketing of farm machinery, private traders were allowedto import farm machinery at zero tariff, which made the price of farm machines effec-tively low. Moreover, incentives and assistance were afforded to local manufacturersof power tillers, seeders, weeders, dryers, and threshers to make farm machinery moreaccessible to farmers. Despite the presence of public and private machinery stations,few farmers mechanized their farm operations. The national average for mechanizedland preparation was estimated to be 9% of the total rice area, and only about 1% ofthe total harvest was threshed mechanically.

Ownership of agricultural machines in the study area was limited to the irriga-tion pump, power tiller, and rice mill (Table 10). The most common farm machineryfound in the villages was the irrigation pump. About 12–16% of the households ownedat least one pump for rent or personal use. Irrigation pumps in the rainfed village werebeing rented to farmers from the nearby irrigated villages. Power tillers and rice millswere only present in the villages practicing intensive rice cultivation. The percentageof households owning such machines, however, was very low.

The only farm operations observed to be mechanized were land preparationand threshing. The percentage of farmers using tractors for land preparation and me-chanical threshers for threshing, however, was relatively small compared with thepercentage of farmers using manual operations. This was attributed to the high cost ofrental fees for these machines. Tractors were normally rented together with an opera-

420 Garcia et al

tor for $5 per man-machine day. On the other hand, a pair of bullocks with an operatorcan be hired for only $1.65 per man-animal day. The rent for the thresher was deter-mined by the volume of paddy threshed. The average rental fee was $0.05 per basketor $2 per ton of paddy threshed. Irrigation pumps were rented on a daily basis at therate of $5 per day with the cost of diesel ($2.50 per gallon) being shouldered by thefarmers.

Fertilizer useThe fertilizer supply in the country was mostly procured through importation. Riceproduction consumes 95% of the total available fertilizer (MAS 1996). Prior to 1988,the government directly controlled the importation and distribution of agrochemicals,including fertilizers, to the farmers. Because of the country’s constraints in foreignexchange earnings, however, fertilizer importation had always been low and insuffi-cient. Farmers commonly used urea, triple superphosphate (TSP), and muriate ofpotash (MP). About 72% of the total fertilizer imports were in the form of nitrogenfertilizer (urea).

Total fertilizer use increased steadily from 1.9 kg ha–1 in 1963 to a peak of 72kg ha–1 in 1984. This dramatic growth in fertilizer use was due to the government’smassive fertilizer subsidy that artificially lowered fertilizer prices by 50%. As a re-sult, a significant increase in rice yield was achieved during this period. After 1985,however, fertilizer use continuously declined, up to a low of 21 kg ha–1 in 1992,because of the foreign exchange constraint that prevented the government from sup-plying enough fertilizer to the farmers. In 1993, however, the government privatizedthe marketing and distribution of fertilizer, which made it more available to farmers.Hence, fertilizer use started to pick up and reached 50 kg ha–1 in 1996. Nevertheless,fertilizer was still grossly underused on most farms due to its limited access and highcost. It was estimated that farmers were actually applying only 40% of the recom-

Table 10. Ownership of agricultural machinery and equipment,Nyaungdon Township, 1996.

IrrigatedAgriculturalmachines Rainfed Lower Upperand equipment lowland terrace terrace Deepwater

(% of HHa owning)

Tractor – – – –Power tiller – 6 2 4Irrigation pump 12 12 15 16Threshing machine – – – –Winnower/blower – – – –Rice huller – – 2 3Animal cart 60 41 35 20Fodder cutter – – – –

aHH = households.


mended level of fertilizer use (120 kg urea, 60 kg TSP, and 60 kg MP ha–1 or 55 kg N,27 kg P2O5, and 36 kg K2O ha–1 (MAI 1995).

In the study area, fertilizer use among farmers practicing intensive rice cultiva-tion was higher than that of the rainfed farmers, but still far below the recommendedlevel. On average, farmers applied a similar amount of fertilizer N (about 32 kg ha–1)to both traditional and modern varieties (Table 11). This amount fell below the rec-ommended N rate by 40%. Farmers using modern varieties, however, applied moreP2O5 and K2O in the form of triple superphosphate and muriate of potash, respec-tively.

The use of farmyard manure (FYM), on average, was higher among rainfedfarmers (3,068 kg ha–1) than among the progressive farmers (2,331 kg ha–1). Thisimplied that the rainfed farmers tried to compensate for the deficiency in commercialfertilizer by using more organic fertilizer because of its lower price.

CreditCredit in the villages was obtained both formally and informally. Since no banksexisted in the villages, formal loans could be obtained only from the township’s Ag-ricultural Development Bank. Crop loans and other medium- to long-term loans wereavailable to farming households only. Loans were extended to farmers in groups (five

Table 11. Fertilizer use (kg ha–1) under modern versus traditional varieties by farming practicesand season, Nyaungdon Township, Myanmar, 1996.a

Rainfed Irrigated(one rice cropping) (double cropping)

Fertilizers used Wet season Wet season Summer rice

Traditional Modern Traditional Modern Traditional Modern

Urea 35 (16) 28 (13) 41 (19) 51 (24) 62 (29) 65 (30)Triple superphosphate 5 (2) 30 (13) – 71 (32) – 65 (29)Muriate of potash 16 (9) 13 (8) – – – 31 (19)Farmyard manure 3,272 2,863 2,045 3,272 818 2,045

Double monsoon rice Rice-fish culture

Wet Wet Wet DryFertilizers used season 1 season 2 season season

Traditional Modern Traditional Traditional Modern Modern

Urea 80 (37) 76 (35) 111 (51) 94 (43) 79 (36) 115 (53)Triple superphosphate 25 (11) 30 (14) 65 (29) 62 (28) 92 (41) 89 (40)Muriate of potash – – – 41 (25) – 72 (43)Gypsum – – – – – 124Farmyard manure 2,863 3,272 5,726 818 1,636 818

aNumbers in parentheses are in kg N, P2O5, or K2O ha–1.

422 Garcia et al

in a group) at the rate of $60 per ha in areas under intensive cropping and $20 per hain rainfed areas for a duration of 6 months.

On average, the amount of the crop loans observed in the study area rangedfrom $19 to $24 per farmer (Table 12). On the other hand, the amount of medium- tolong-term loans ranged from $20 to $812. These government-extended loans weredesigned to help farmers procure their farm inputs (e.g., fertilizer and seeds) and rentfarm equipment such as tractors and irrigation pumps. The interest rate for govern-ment loans was remarkably low at 1.5% per month or 18% per annum.

Among the landless households, loans were available only from informal sourcessuch as moneylenders, traders, shopkeepers, relatives, and friends. A majority of theloans were obtained from moneylenders (56%). Friends and relatives were also com-mon sources of credit (41%). Interest rates varied considerably from 5% to 12% permonth or 144% per annum. Landholders also obtained loans from informal sourcessince government loans could only be used to finance farm-related expenses. Hence,credit for personal needs was obtained mostly from informal channels. Occasionally,the school’s cooperative operated by the PTA (parent-teacher association) extendedloans for personal needs of the villagers.

Larger loans were observed among farm households belonging to the irrigatedvillages. This can be related to intensive rice cultivation, which required more cashinputs. Such loans were obtained from both formal and informal sources dependingon the size of the farmer’s landholding. Generally, the larger the farmer’s landhold-ing, the larger the loanable amount, since landholding size can be a good indicator ofthe farmer’s capacity to pay back loans. Notably, in the deepwater village where rice-fish farming was promoted by the government, more credit was extended to farmersto enable them to invest in rice-fish culture.

The average loans of landless households ranged from $2.50 to $67. A majorityof the loans taken out by landless households were spent to finance businesses (about70–75%) involving home gardening, livestock raising, and other nonfarm activities(Table 13). Among the personal expenses of the landless that were often financed byborrowing were health care (6–14%) and religious ceremonies (5–9%). The samepattern of use of loans obtained informally was observed among the farming house-holds. Despite the presence of government credit, a majority of the informal loans(53%) were spent to finance agricultural investments. This indicates that credit ex-tended by the government to farmers was still inadequate.

Marketing and prices of rice outputBefore 1987, the government had always strictly regulated the rice market in Myanmar.Aside from home consumption and seed requirements, all of the farmers’ rice outputhad to be sold to the state economic enterprises and cooperatives at predeterminedprices. Immediately after the 1988 crisis, the production quota was removed and farmerswere able to sell their produce privately at prevailing market prices. In 1990, how-ever, the production quota was revived but was reduced to 593 kg ha–1. Surplus pro-duction was privately marketed. In depressed areas, the quota was lower dependingon the existing rice yields in the locality. Only the harvest from the monsoon rice crop


Tabl

e 1

2. S

ourc

es a

nd c

osts

of lo

ans

by s

ize

of la

ndho

ldin

g in

Nat

pay

(rai

nfed

) vi

llage

.

Rai

nfed

Irrig

ated

Land

less

Land

owne

rLa

ndle

ssLa

ndow

ner

Sou

rces

% o

fAv

erag

eAv

erag

e%

of

Aver

age

Aver

age

% o

fAv

erag

eAv

erag

e%

of

Aver

age

Aver

age

HH

aam

ount

inte

rest

HH

amou

ntin

tere

stH

Ham

ount

inte

rest

HH

amou

ntin

tere

stof

loan

sra

teof

loan

sra

teof

loan

sra

teof

loan

sra

te(U

S$)

(%)

(US

$)

(%)

(US

$)

(%)

(US

$)

(%)

Form

al/b

anks

C

rop

loan

––

–7

51

91

.8–

––

51

24

1.4

M

ediu

m-te

rm lo

an–

––

1

20

1.5

1

67

1.5

2

8

12

1.2

L

ong-

term

loan

––

––

––

––

–

2

319

4.4

S

choo

l coo

pera

tive

3

17

0

.8

13

80

.8–

––

––

–In

form

al

Mon

ey le

nder

29

48

12

.0

83

27

.95

36

11

01

7

10

07

.8

Tra

ders

3

2.5

0–

––

––

–

13

,39

66

.0

Frie

nds/

rela

tives

66

42

12

.31

55

06

.2

54

0

4

.92

7

19

87

.5

a HH

= h

ouse

hold

s.

424 Garcia et al

was subjected to the production quota system. The farmers could privately market allof the harvest from summer rice cropping. This policy was implemented in 1992 toencourage farmers to plant summer rice and boost rice production in irrigated areas.

The government’s procurement price had always lagged behind the market price,from 38% in 1963 to a maximum of 65% in 1986 (Table 14). As retail prices started tosoar in the 1990s, the government was forced to adjust the procurement price, whichnarrowed the gap to about 17% until 1995. At the village level, however, results of thestudy showed that the government’s procurement price lagged behind the market priceby as much as 69% in the same year (Table 15).

In 1996, the government again abolished the production quota, which allowedfarmers to sell all their produce privately. After just two months of implementation,however, the government decided to put back the quota system but adjusted the pro-curement price by 300% (from $30 t–1 to $125 t–1). Despite this effort, the government’sprocurement price still lagged behind the market price by as much as 62%, i.e., thecurrent market price was $333 t–1.

The quota system of the government in effect created an indirect tax that penal-ized the rice farmers. Results of the study showed that about 27% of the farmers’ totalharvest was sold to the government as a production quota. The average tax per farmerwas therefore valued at $85, assuming that the quota could be sold privately at marketprices. In turn, this implicit tax was estimated to be 7% of the farmers’ total farmincome. The tax burden of the production quota became more acute for farmers withsmaller landholdings, for whom the proportion of the production quota to total har-vest can go as high as 63%. In some unfortunate circumstances, farmers needed tobuy paddy from the market to fulfill their quota for the cropping season.

Table 13. Pattern of loan use (% of loan spent), Nyaungdon Township, 1996.

Rainfed Irrigated

Uses Landless Landowner Landless Landowner

Formal Informal Formal Informal Formal Informal Formal Informal

InvestmentAgric. fixed investment – – – – – – – 12Agric. current expenses – 26 27 15 75 9 13 2Livestock business – 15 5 16 – 55 83 64Non-agric. business – 29 56 39 – 11 – 17

ConsumptionFood 100 6 4 7 25 4 2 1Education/health – 14 8 14 – 6 – 1Housing – 3 – – – 6 – 1Social/religious – 6 – 9 – 5 – 1Others – – – – 5 2 1


Paddy output was often transported to the government’s production camps (about2 km from the villages) using bullock carts and tractors. Other means of transportwere trucks, boats, bicycles, and carrying on the head. On the other hand, paddyoutput marketed privately was generally picked up by rice traders at the farm gate.

Table 14. Government procurement price and market price of ricefrom 1963 to 1995, Union of Myanmar.

Government Market price Price gap DifferenceYear procurement price (kyats t–1) (kyats) (%)

(kyats t–1)

1963 239 385 146 381971 276 432 156 361976 695 1,174 479 411979 695 1,286 591 501984 695 1,645 950 581985 732 1,753 1,021 581986 732 2,082 1,350 651987 1,774 2,040 266 131988 3,075 3,717 642 171989 4,331 4,380 49 11990 4,391 4,638 247 51991 4,198 5,371 1,173 221992 5,924 8,642 2,718 311993 9,473 12,164 2,691 221994 12,555 15,563 3,008 191995 14,676 17,746 3,070 17

Sources: Data from 1963 to 1992, Hossain and Marlar Oo (1995). Data from1993 to 1995 (Statistical Yearbook 1995, Ministry of National Planning andEconomic Development (1995).

Table 15. Marketing and disposal of rice production, Nyaungdon Township, 1996.

IrrigatedAverage

Variables Rainfed Lower Upper Deepwater allterrace terrace villages

Total amount produced (baskets) 208.68 220.66 217.17 278.94 231.36Amount sold to government (baskets) 56.01 51.21 60.77 82.53 62.63Price from sale to government (kyats) 74.02 73.06 73.45 72.98 73.38Amount sold to market (baskets) 69.21 86.65 141.46 211.81 127.28Price from sale to market (kyats) 221.31 245.56 232.00 246.30 236.29Average distance to government sales 5.92 5.05 5.55 5.36 5.47

center (km)Average distance to market center (km) 1.61 1.82 1.69 1.75 1.72

426 Garcia et al

Labor and employmentTable 16 presents the occupational distribution of the economically active populationin the villages from 16 to 59 years old. A majority of the people (64–69%) wereengaged in agriculture as farm operator, hired laborer, or livestock holder. The secondlargest population group was composed of dependents, that is, housewives and mi-nors (18–25%). Students represented 3–5% of the population. About 5% of the popu-lation was engaged in trading and market vending. The group of government employ-ees working as teachers, clerks, and extension workers was less than 1%. Job oppor-tunities in rural industry, construction, and transport services were very low, averag-ing about 1% for each employment group.

Household members with a farm operator as head of the household with extratime from their farms also worked in a secondary occupation as hired laborers. Womenkept themselves busy after working on their farms by tending vegetables, flowers,and betel leaves in their homegardens to augment family income. Taking care of smalllivestock such as chickens, ducks, and pigs also provided a secondary livelihood.

Table 17 presents the labor participation of children 10 to 15 years old in thefour villages. The proportion of children who were participating in economic activi-ties at an early age ranged from 25% to 34%. A majority (66–75%) of the childrenwere still attending school in this age bracket. Notably, a higher proportion of femalechildren quit school early to participate in economic activities (32–40%) than malechildren (15–35%) because of the scarcity of intermediate schools in the villages.Hence, only male children who reached high school were often sent to the townshipto study. After finishing elementary school, female children were generally commis-sioned to help their parents in their livelihood.

Labor for seasonal agricultural work was generally supplied by the landlesshouseholds. They spent about 160–190 days per year working as farm laborers (Table18). Although landowners participated in the agricultural labor market, the averagenumber of days they spent as laborers was 66% lower than that of the landless. Theparticipation rate decreased as the size of the landholding increased. Notably, thedemand for agricultural laborers was 15% higher in the villages where intensive ricecultivation was being practiced than in the rainfed village. This was due to the shortturnaround time required in intensive rice cultivation; hence, more laborers were neededto finish the farm tasks.

Working as farm help also provided the landless with a steady job. Generally,young boys between 12 and 16 years old were hired as farm help by landowners totake care of their livestock. In the rice-fish area, many landless laborers worked ascaretakers of fishponds and as seasonal laborers during fish harvesting. Transportoperation (pedicabs and boats), collection of fuelwood, personal service, andshopkeeping also kept the landless households busy, especially during lean monthswhen farm work was scarce. Industrial jobs were not available in the villages. Indus-trial workers often needed to seek employment in the townships or cities. The numberof days spent by the landless for all types of jobs was generally higher than that of thelandowning households. Because family members of farm households were often


Tabl

e 1

6. Em

ploy

men

t an

d in

com

e fr

om n

onfa

rm a

ctiv

itie

s, N

yaun

gdon

Tow

nshi

p, 1

996.

Irrig

ated

vill

ages

Rai

nfed

Low

er t

erra

ceU

pper

ter

race

Dee

pwat

er

Non

farm

sou

rces

Aver

age

Aver

age

Aver

age

Aver

age

Aver

age

Aver

age

Aver

age

Aver

age

days

of

annu

alda

ys o

fan

nual

days

of

annu

alda

ys o

fan

nual

empl

oym

ent

inco

me

empl

oym

ent

inco

me

empl

oym

ent

inco

me

empl

oym

ent

inco

me

(kya

ts)

(kya

ts)

(kya

ts)

(kya

ts)

Agric

ultu

ral l

abor

27

71

8,2

39

32

92

2,0

68

36

02

6,1

72

29

21

9,9

89

Indu

stria

l lab

or1

58

9

,38

62

05

16

,79

61

09

15

,88

02

41

83

,83

5C

olle

ctio

n of

fire

woo

d/fu

el/

12

11

3,0

38

40

01

5,8

00

––

26

42

6,4

00

fore

stry

Anim

al h

usba

ndry

46

51

4,0

55

––

9

0

2,2

50

30

0

9,9

00

Fish

ing

17

21

1,5

50

9

9

9,3

92

14

81

9,2

67

16

51

3,4

83

Trad

ing/

shop

keep

ing

20

32

2,9

00

19

01

5,1

91

22

92

3,9

54

20

44

1,3

29

Tran

spor

t op

erat

ion

24

82

3,2

11

22

21

7,9

23

28

23

5,9

65

17

04

8,7

50

Con

stru

ctio

n la

bor

14

21

5,7

56

10

51

2,1

67

8

5

6,1

44

19

21

9,7

92

Pers

onal

ser

vice

31

83

3,9

00

13

4

8,7

57

––

15

8

7,0

12

Gov

ernm

ent

empl

oym

ent

25

7

7,9

80

39

51

0,4

27

36

0

9,8

39

38

12

3,7

50

Ren

tal i

ncom

e3

82

28

,00

01

20

6

,00

01

91

36

,97

31

82

14

,31

0Pe

nsio

n/do

natio

n–

–2

88

17

,28

0–

–1

80

4,5

00

428 Garcia et al

Table 17. Incidence of child labor (age group 10–15 years), Nyaungdon Township,1996.

Irrigated

Child population Rainfed Lower Upper Deepwaterterrace terrace

Males Total number 40 100 83 66 Attending school 80% 65% 76% 85% Participating in economic activities 20% 35% 24% 15%Females Total number 38 74 74 73 Attending school 66% 68% 60% 66% Participating in economic activities 34% 32% 40% 34%All Total number 78 174 157 139 Attending school 73% 66% 68% 75% Participating in economic activities 27% 34% 32% 25%

Table 18. Average per capita employment in nonfarm activities (no. of days peryear), Nyaungdon Township, 1996.

Rainfed IrrigatedNonfarm sources

Landless Landowner Landless Landowner

Agricultural labor 160 58 190 61Collection of fuel 57 90 166 –Animal husbandry 214 270 45 100Fishing 86 – 75 51Industry 74 60 84 46Trading 113 33 116 58Transport 142 53 127 57Construction 79 47 95 31Personal service 154 174 48 86Government employment 288 300 329 268Rental service – 191 106 70

busy with works on their farms, nonfarm activities could be done only during the off-season.

Wages and incomesTable 19 presents the composition of annual household income. Agricultural incomecontributed 51% to total household income while the remaining 49% came from off-farm sources. Income from agricultural labor, about $0.50 per man-day, contributed10% of the total income. Other agricultural income generated by households mostlycame from farming, home gardening, and raising small livestock in the backyard.


Table 19. Composition of household annual income (% of total income), NyaungdonTownship, 1996.

Rainfed IrrigatedSources Average

Landless Landowner Landless Landowner

AgricultureFarm/garden 7 66 4 37 28Agricultural labor 11 3 14 10 10Animal husbandry 6 3 1 1 3Fishing 6 – 9 4 5Forestry 7 2 9 – 5

Total 37 74 37 52 51NonagricultureIndustry 8 2 9 10 7Trading 14 3 14 16 12Transport 15 1 9 4 7Construction 9 2 11 2 6Personal service 12 10 2 1 6Gov’t employment 5 4 6 4 5Rental service – 4 9 4 4Transfer – – 4 – 1

Total 63 26 64 41 49

A majority of the income (64%) of the landless households was generated fromnonfarm activities. Trading, transport operation, personal service, and constructionlabor were the major sources of nonfarm income, with shares of about 9–15% each intotal income.

On the other hand, a larger proportion (63%) of the total income generated bylandowner households came from agricultural activities. The higher income of theirrigated farm households compared to the rainfed households was attributed to theintensive rice cultivation in these villages. Income from agricultural labor contrib-uted a higher percentage in the irrigated villages (10% vs 3%) because of the in-creased opportunity for farm work brought about by the new rice-based productionsystems. Trading, transport operation, and personal service were also the major sourcesof nonfarm income for the farm households, which contributed 34% to total income.A higher share of nonfarm income in the farming households belonging to the irri-gated villages (41% vs 26%) was observed, implying that family members were ableto engage in more high-paying off-farm activities since farm operators tended to hiremore agricultural laborers to do the farm tasks.

Income distributionTo determine the pattern of income distribution in each village, the households wereranked with respect to per capita income and the corresponding income shares of

430 Garcia et al

Table 20. Gini ratios for total income and its components, Nyaungdon Township,1996.

IrrigatedDistribution Rainfed

Lowland Upland Deepwater All

IncomeAgricultural labor income 0.69 0.68 0.67 0.75 0.70Farm income 0.59 0.72 0.79 0.59 0.68Nonfarm income 0.80 0.75 0.81 0.84 0.81Total income 0.35 0.41 0.48 0.47 0.44Ownership of resourcesLand 0.66 0.68 0.68 0.56 0.65Education 0.25 0.28 0.22 0.24 0.25

Fig. 1. Lorenz curves of per capita total income.

successive decile groups were estimated. Figure 1 shows a series of Lorenz curves1

depicting the pattern of income distribution in the villages where different rice pro-duction systems were being used.

To infer the degree of inequality in income distribution, the Gini concentrationratio2 was estimated based on per capita household income. Table 20 presents resultsof the analysis. The Gini ratio of income generated from all sources, such as farm,

1The Lorenz curve shows the percentage share of income received by different population groups. It shows thedegree of equality and inequality in income distribution. The greater the departure of the Lorenz curve from the45° line, the more unequal the distribution of income.2The Gini concentration ratio is the estimated area inside the Lorenz curve. The value of the ratio ranges from0 to 1. The closer the value of the Gini ratio to 1, the more unequal the distribution of income.

1.0

0.8

0.6

0.4

0.2

0.00.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Cumulative proportion of population

Cumulative proportion of income

45 degree lineRainfedLower terraceUpper terraceDeepwaterAll villages


agricultural labor, and nonfarm income, was 0.44 for all the villages. The rainfedvillage estimate (0.35) was lower than that of the irrigated villages (ranging from 0.41to 0.48). This indicates that the distribution of income in the rainfed village was lessunequal compared with that of the other villages with access to irrigation facilities,which enabled intensive cropping.

The estimated Gini ratio for farm income registered an average of 0.68 for allvillages. The ratio was expected to be significantly higher than the aggregate incomedue to the highly unequal distribution of land ownership in the villages. The Giniconcentration ratio for land ownership was estimated at 0.65. Notably, in the deepwaterarea where common lands were still available for public cultivation, the Gini ratiowas lower (0.56). The distribution of farm income at the village level showed inter-esting Gini coefficients. The rainfed and deepwater areas registered lower ratios (both0.59) than the villages located in the upper and lower terraces (0.79 and 0.72, respec-tively). This observation further emphasized that the presence of irrigation facilitiescontributed to widening the disparity of farm income among landowners.

The average Gini ratio for agricultural labor income in all four villages wasestimated to be 0.70. However, the deepwater village showed an even higher ratio(0.75) than the other villages (ranging from 0.67 to 0.69). This showed that the oppor-tunity for family labor (especially for the landless) to work in the fields as hiredlaborers was hampered by excessive flooding in the area during the monsoon season.

The highest Gini ratio was registered by nonfarm income (0.81 for all villages).This indicates that greater income inequality resulted from the differential access ofhousehold members to nonfarm jobs. This may be due to fewer opportunities fornonfarm activities available in the villages or a lack of the skills required in mostnonfarm jobs. Gini ratios for the distribution of education were likewise generated todetermine how individuals in the villages differed in per capita access to education.Estimates showed that the ratios were generally low (0.22 to 0.28), indicating thatindividual members of the active population had relatively the same level of educa-tion across villages (at most 7 years of schooling).

Cost and return analysis of different rice production systems

To determine the profitability of the different rice-based technologies promoted bythe government in the study area, a cost and return analysis was done for each farmcategory per season (Table 21). Values for cost and return components were expressedon a per hectare basis. The construction of the cost accounts in the analysis was basedon the imputation of existing wage rates and rental fees for all activities using familylabor and self-owned resources. At the same time, paddy prices (both governmentand market prices) were used to value the total harvest. The net return should there-fore be seen as an indicator of the farmer’s ability to recover both cash and noncashcosts of rice production instead of the conventional idea of profit.

Results of the study showed that net gain from the different rice productionsystems ranged from $30 to $95, whereas net loss ranged from $48 to $135. How-ever, if production cost were computed on the basis of pure cash costs (i.e., not valu-

432 Garcia et al

Tabl

e 2

1. C

ost

and

retu

rn a

naly

sis

(val

ues

in k

yats

) of

ric

e cr

oppi

ng u

nder

diffe

rent

pro

duct

ion

syst

ems,

Nya

ungd

on T

owns

hip,

1996.

Dou

ble

mon

soon

aD

oubl

e cr

oppi

ngR

ice-

fish

Item

Rai

nfed

WS

1W

S2

DS

WS

DS

WS

DS

Cos

tsM

ater

ial c

ost

5,1

05

12

,27

01

4,7

88

16

,70

71

5,2

26

16

,70

71

6,4

49

22

,44

7La

bor

cost

13

,48

31

2,6

55

14

,16

22

1,1

70

30

,32

02

1,1

70

14

,19

82

0,9

16

T

otal

cos

t1

8,5

88

24

,92

52

8,9

50

37

,87

71

5,0

94

37

,87

73

0,6

47

43

,36

3R

etur

nsM

arke

ted

A

mou

nt2

,51

87

69

3,3

33

1,7

24

96

61

,72

41

,11

72

,96

3

Pric

e1

0.0

01

1.4

01

1.1

01

1.8

01

1.1

01

1.8

01

0.4

51

2.0

0Q

uota

A

mou

nt5

93

59

3–

–5

93

–5

93

–

Pric

e3

.75

3.7

5–

–3

.75

–3

.75

–S

ale

from

ric

e st

raw

s8

45

44

83

,33

42

,83

81

,15

42

,83

86

07

2,0

56

T

otal

ret

urns

22

,21

91

1,4

39

40

,33

02

3,1

81

14

,10

12

3,1

81

14

,50

43

7,6

12

Net

ret

urns

sea

son–

13

,73

1(1

3,4

86

)1

1,3

80

(14

,69

6)

(16

,21

9)

(14

,69

6)

(16

,14

3)

(57

5)

N

onca

sh c

osts

7,7

55

6,2

46

7,6

85

8,8

47

8,1

87

8,8

47

7,0

48

9,2

56

Ret

urn

over

cas

h co

st1

1,8

46

(7,2

40

)1

9,0

65

(5,8

49

)(8

,03

2)

(5,8

49

)(9

,09

5)

3,5

05

a WS

= w

et s

easo

n, D

S =

dry

sea

son.


ing self-owned resources), net gain resulted in a wider range ($29 to $160) but with ahigher frequency of occurrence since net losses were minimized. This observationcan be attributed to the significant proportion of noncash costs with respect to totalcost (29%).

In general, rice monocropping in the rainfed village generated positive net re-turns compared with intensive rice cultivation. The lower net returns in the irrigatedvillages was due to the high cost of irrigation for dry-season cropping, that is, the highcost of diesel fuel and pump rental. Likewise, farmers engaged in intensive cultiva-tion used more fertilizers, hence the higher production costs. Despite the higher fertil-izer use observed on intensive rice farms, however, yields did not increase enough tocover the high production costs since fertilizer use was still inadequate.

Another important factor that must be considered in the farmers’ failure to re-cover their production costs was their inability to cope with the technological andeconomic demands of the new rice-based cropping systems in terms of time, labor,and capital resources associated with intensive rice cultivation. Note that among thethree programs advocated by the government (summer rice cropping, double mon-soon cropping, and rice-fish farming), only the summer rice production showedsustainability in farmer adoption. Double monsoon cropping was totally abandonedafter one year of implementation. Likewise, adopters of rice-fish farming drasticallymodified the system in favor of fish culture.

Major production constraints

To summarize the impacts of biophysical and socioeconomic factors on the rice pro-duction systems in the country, major production constraints (both biotic and abiotic)were identified and their effects quantified using the yield gap method. Through thistechnique, normalized yield losses associated with these constraints could be assessedand evaluated to determine their relative importance. Results of the study showed thatthe abiotic constraints significantly affected a larger percentage of the total rice areaand the estimated yield losses were higher compared with those of the biotic con-straints (Tables 22 and 23).

Biotic constraints to rice productionThe biotic constraints commonly reported by farmers in the survey area can be di-vided into three major categories: genetic problems, insect pests, and diseases. Ge-netic problems included varietal degeneration, stunted growth, and lodging. Resultsof the study showed that the major genetic problem was varietal degeneration. Theestimated normalized yield loss associated with this problem was about 31 kg ha–1,resulting in a projected total production loss of 181,327 tons or 1% of the total na-tional production. The primary reason for this problem was the continuous use ofseeds coming from the farmer’s own harvest without proper knowledge of the tech-niques for seed selection.

For insect pests, seven species were identified by farmers in the survey area:rice caseworm, rice hispa, rice stem borer, rainball, gall midge, armyworm, and brown

434 Garcia et al

planthopper. Among these insect pests, the rice caseworm was considered the mostdestructive, resulting in an estimated yield loss of 8 kg ha–1. This figure, however,translates into only 0.27% of total national production. The total damage caused byall these insects combined was estimated to be less than 1% of the total nationalproduction.

Although insect pests were commonly observed on rice farms, their occurrenceand infestation were still low and considered to be insignificant. Because of the lim-ited supply and high cost of agrochemicals, most farmers did not apply insecticides.This resulted in the nondestruction of beneficial organisms, which actually helped inthe control of harmful insects.

Rice disease in the country is a new phenomenon and was detected only in thelate 1980s. The major diseases that were increasingly affecting Myanmar rice pro-duction were bacterial leaf blight, blast, and bacterial leaf sheath. Bacterial leaf blightcontributed the highest damage incidence in terms of total national production loss(0.21%) with a normalized yield loss of 6 kg ha–1.

Based on the survey, the farmers reporting problems of rice diseases were mostlyusers of modern varieties, especially in the dry season. Farmers were generally un-aware of the causes and treatment of these diseases. Stubbles in the infected fields

Table 22. Estimated rice yield and production losses due to biotic constraints in Myanmar,1997.

% of total % Estimated Normalized Estimated % loss withBiotic rice area occurrence yield yield loss production respect toconstraints affected in 10 years loss (kg ha–1) loss (Mt) national

(kg ha–1) production

GeneticVarietal degeneration 8.59 80 454 30.97 181,327 1.05Stunted growth 1.41 73 707 7.24 42,410 0.25Lodging 0.72 80 363 2.08 12,174 0.07

Total 235,910 1.36

Insect pestsCaseworm 3.76 84 251 7.92 46,364 0.27Rice hispa 0.85 31 468 1.23 7,200 0.04Stem borer 0.74 69 251 1.29 7,552 0.04Rainball 0.27 55 311 0.46 2,693 0.02Gall midge 0.21 15 482 0.15 878 0.01Armyworm 0.21 28 492 0.28 1,639 0.01Brown planthopper 0.03 10 1,374 0.04 234 0.00

Total 66,560 0.38

DiseasesBacterial leaf blight 2.80 56 398 6.20 36,295 0.21Blast 0.28 55 1,749 2.65 15,513 0.09Bacterial leaf sheath 0.47 65 246 0.75 4,391 0.03

Total 56,198 0.32


Tabl

e 2

3.

Esti

mat

ed r

ice

yiel

d an

d pr

oduc

tion

loss

es d

ue t

o ab

ioti

c co

nstr

aint

s in

Mya

nmar

, 1997.

Abio

tic%

of to

tal r

ice

% o

ccur

renc

eEs

timat

ed y

ield

Nor

mal

ized

Estim

ated

nat

iona

l%

loss

cons

trai

nts

area

aff

ecte

dlo

ss (

kg h

a–1)

yiel

d lo

sspr

oduc

tion

loss

with

res

pect

(kg

ha–

1)

(Mt)

to n

atio

nal

prod

uctio

n

Agro

nom

icIn

suff

icie

nt u

se o

f ch

emic

al fer

tiliz

er6

38

38

97

47

4.5

2,7

77

,65

21

6.0

6Po

or w

eed

man

agem

ent

24

60

66

89

4.6

55

3,4

75

3.2

0Po

or w

ater

man

agem

ent

15

33

81

24

0.5

23

6,8

16

1.3

7U

se o

f ol

d se

edlin

gs6

57

59

51

8.7

10

9,4

37

0.6

3To

tal

3,6

77

,38

02

1.2

6

Clim

atic

/env

ironm

enta

lFl

ood

2.7

36

44

87

8.5

49

,87

60

.29

Dro

ught

at

vege

tativ

e st

age

1.2

53

87

94

3.7

21

,83

50

.13

Dro

ught

at

seed

ing

1.1

74

21

,19

25

.93

4,4

22

0.2

0D

roug

ht a

t an

thes

is0

.30

51

1,1

77

1.8

10

,65

40

.06

Col

d at

ant

hesi

s0

.22

54

1,1

92

1.5

8,4

88

0.0

5To

tal

12

5,2

75

0.7

3

Soi

l-rel

ated

Acid

ic s

oils

4.1

51

74

04

2.8

16

,39

10

.09

San

dy s

oils

3.9

94

09

29

14

.88

6,7

56

0.5

0Lo

w f

ertil

ity3

.90

30

1,1

69

13

.78

0,0

83

0.4

6Al

kalin

ity0

.86

27

1,1

99

2.7

16

,04

00

.09

Sal

inity

0.5

813

486

0.4

2,2

25

0.0

1To

tal

20

1,4

95

1.1

5

Soc

ioec

onom

icH

igh

cost

of

fert

ilize

r4

9.4

06

48

60

27

1.6

1,5

89

,82

99

.19

Unt

imel

y an

d in

adeq

uate

sup

ply

0.8

04

01

,81

45

.83

3,6

61

0.1

9of

fer

tiliz

erLa

ck o

f cr

edit

faci

lity

46

.80

55

66

61

70

.81

,00

0,0

97

5.7

8Lo

w a

dopt

ion

of n

ew t

echn

olog

y1

9.1

03

67

96

54

.33

17

,75

51

.84

Sea

sona

l lab

or s

carc

ity1

0.5

07

02

33

17

.21

00

,45

50

.58

Lim

ited

tech

nica

l ext

ensi

on s

ervi

ces

6.2

02

19

45

12

.06

9,9

55

0.4

0Ir

rigat

ion

faci

lity

dete

riora

tion

5.2

01

96

28

6.3

36

,58

80

.21

Inad

equa

te s

uppl

y of

goo

d se

ed0

.30

25

77

70

.52

,98

60

.02

Tota

l3

,15

1,3

25

18

.21

436 Garcia et al

were often burned for disease control. The use of fungicides was almost nil due to thefarmers’ limited knowledge of their use and the high cost of these chemicals.

Abiotic constraints to rice productionThe abiotic constraints reported by farmers in the survey area were categorized intothe following: agronomic problems, climatic and/or environmental factors, soil-re-lated problems, and socioeconomic factors. Four major agronomic problems in riceproduction were identified: insufficient use of chemical fertilizer, poor weed and watermanagement, and the use of old seedlings. Of the four problems, the most serious wasinsufficient use of chemical fertilizer. It was reported in 63% of the total rice areasurveyed, with a normalized yield loss of 474 kg ha–1. This translates into a totalproduction loss of about 2.8 million tons or 16% of the total national production.Because of the high cost of commercial fertilizer, farmers were observed to apply lessthan half of the recommended rate of nitrogen fertilizer and almost none of the phos-phorus and potassium fertilizers.

Other agronomic constraints caused by poor weed and water management re-sulted in normalized yield losses of 95 and 40 kg ha–1, respectively, with estimatednational production losses of 3% and 1%. Weed problems were often attributed tofarmers’ inability to prepare and level rice fields properly, which made weeding diffi-cult. Also, the practice of broadcasting as a method of crop establishment aggravatedthe weed problems. At present, the use of herbicides is minimal due to the farmers’lack of knowledge about the use of chemical weed control and the high cost of herbi-cides.

On the other hand, poor water management was often associated with farmers’inability to control the water supply because of insufficient water from rainfall andirrigation systems. In central and upper Myanmar, irrigation facilities were constructedto supplement rainfall for the wet-season crop since water was always insufficient inthe dry zone. However, despite the government’s effort to boost water supply in theseareas, water stress from drought still posed an important constraint in the dry zone.

Another important abiotic constraint was brought about by socioeconomic fac-tors. The two most important socioeconomic constraints reported by more than 80%of the township surveyed were the lack of credit facilities (87%) and the high cost ofcommercial fertilizers (83%). These constraints were reported to bring about normal-ized yield losses of 171 and 272 kg ha–1, respectively. Production losses were esti-mated to be 6% and 9% of the national production, respectively. These problemswere related, however, since credit was normally used to purchase agricultural inputs,especially chemical fertilizer. With the high cost of fertilizer coupled with the lack ofcredit facilities, most farmers could barely afford to purchase the much-needed amountof fertilizer.

Other socioeconomic constraints reported by farmers were low adoption of newrice technologies, seasonal labor scarcity, limited technical extension services, dete-rioration of irrigation facilities, and inadequate supply of good seed material. Theeffects of these constraints on national production, however, were minimal, reachinga maximum of only 2%. Likewise, stresses brought about by climatic/environmental


and soil-related problems contributed low production losses and were both estimatedto be about 1% of total national production.

Conclusions

Myanmar has a substantial potential for increasing rice production. A wide range ofpolicy options is available to policymakers to tap this potential. The implementationof the summer rice (dry-season) program in 1992, coupled with irrigation infrastruc-ture development, successfully showed that this potential could easily be exploited.Unfortunately, given the present coverage of irrigation facilities (only 26% of thetotal rice area), this potential has not been maximized, despite the availability of abun-dant water sources for irrigation. More infrastructure development is needed to movethe entire country into rice double cropping. Low-cost irrigation structures such asshallow tube wells (a technology still unknown to Myanmar farmers) can be har-nessed to provide the much-needed water for dry-season production.

Furthermore, reclamation of the culturable wastelands and fallow lands in thedelta area and elsewhere proved to be a strategic policy for extending the land frontierfor rice and other crop production. The recent land development policy of the govern-ment for the large corporations operating in the country, which aimed to encourageagricultural production and exports, may prove to be ideal along this line. However,extreme caution should be exercised in monitoring the activities of these corporationsto ensure that increased agricultural production is indeed achieved, instead of simplyrechanneling the existing domestic output to international markets.

Increasing yield is also one of the potential sources of rice output growth inMyanmar. One important factor that constrained rice yields, however, was the inad-equate use of fertilizer. It was observed that farmers applied only about 40% of therecommended fertilizer rate for both modern and traditional varieties. This was pri-marily due to the high price of commercial fertilizer and its limited supply in thevillages. With approximately 54% of the total rice area grown to modern and im-proved rice varieties, the potential yield of these varieties can only be achieved ifappropriate amounts of fertilizer are applied. The use of FYM helped augment thedeficiency in the use of chemical fertilizer. Despite the use of FYM, however, thetotal available NPK for efficient rice production is still sadly inadequate. There aretwo possible ways to approach the fertilizer problem in the country. First, fertilizerimportation should be deregulated to encourage businessmen to procure it externallyand make it more available to local markets at lower prices. Second, more creditfacilities should be extended to the farmers to enable them to buy and apply the fertil-izer needed to increase their yields.

Proper incentives for farmers can also help increase rice production. The im-plicit tax burden of the production quota can be reduced if the government procure-ment price can be regularly adjusted to reflect market prices.

Landlessness is a pervasive problem among the Myanmar people despite thecountry’s vast land area. This is a critical issue about which policymakers have optedto remain silent. Land distribution in the sample villages was highly unequal, with the

438 Garcia et al

landless population reaching 50–60%. This had a significant effect on the inequalityof income distribution among the village population. Without land, the landless hadlimited sources of livelihood. Their income-generating capacity was severely ham-pered; as a result, poverty among this group was most pronounced.

On the other hand, the landless population provided the much-needed labor forrice production, especially during the peak seasons. Average farm size ranged from0.8 to 2 ha; hence, farm laborers were commonly hired. The potential for mechaniza-tion to solve seasonal labor shortages was encouraging. The cost of buying and/orrenting farm machinery, however, was still prohibitive for farmers.

With the right government policies, incentives, and direction, Myanmar may beable to regain its position as one of the world’s major rice exporters. The potential forincreasing rice production can be achieved with (1) rice planting in the fallow lands(1.3 million ha) coupled with irrigation infrastructure development, (2) the expansionof cultivable land area coming from the available 8.0 million ha of culturable waste-land, and (3) double rice cropping on rainfed lands where abundant water sources canbe tapped for irrigation.

To address the major rice production constraints, Myanmar scientists shouldgive high priority to research on varietal improvement, integrated nutrient manage-ment, integrated pest management, and appropriate agronomic management. Suchresearch is expected to develop production technologies that will overcome presentand anticipated production constraints.

ReferencesHossain M, Marlar Oo. 1995. Myanmar rice economy: policies, performance and prospects.

Paper presented at the Final Workshop on Projections and Policy Implications of Me-dium and Long-Term Rice Supply and Demand Project, Beijing, China.

MAI (Ministry of Agriculture and Irrigation). 1995. Information on Myanmar agriculture. TheGovernment of the Union of Myanmar.

Ministry of National Planning and Economic Development. 1995. Statistical yearbook 1995.The Government of the Union of Myanmar.

Myanma Agricultural Service. 1996. Myanmar agriculture in progress. Ministry of Agricultureand Irrigation, The Government of the Union of Myanmar.

Myanma Agricultural Service. 1997. Rice cultivation situation of Ayeyarwady Division for1996-97. Ministry of Agriculture and Irrigation, The Government of the Union ofMyanmar.

NotesAuthors’ addresses: Y.T. Garcia, Assistant Professor, Department of Economics, College of

Economics and Management, University of the Philippines at Los Baños, Philippines;M. Hossain, Head, Social Sciences Division, International Rice Research Institute, DAPOBox 7777, Metro Manila, Philippines; A.G. Garcia, Representative and Agronomist,IRRI Representative Office, International Rice Research Institute, Yangon, Myanmar.


Acknowledgments: The authors wish to express their sincerest gratitude to the key officials ofthe Myanmar Agriculture Service and Department of Agricultural Planning for theirgenerous support in the conduct of the village survey.


Integration of biophysical and socioeconomic constraints . . . 441

This chapter reviews research incorporating socioeconomic and biophysicalvariables into analysis to characterize rainfed rice environments. The use ofgeographic information systems (GIS) as a tool for integrating these two typesof data is highlighted. GIS starts as a useful tool for organizing and display-ing geo-referenced social and economic data. The increasing ease of use ofGIS software and improvements in geographic positioning systems and re-mote-sensing technology facilitate the generation of more accurate spatiallyreferenced data that can be integrated into analyses of agricultural practicesand outcomes. GIS software, combined with other software, provides analyti-cal techniques for converting distinct types of biophysical and socioeconomicdata to a common scale amenable to analysis as an integrated database.These techniques are outlined. A final area of GIS application in socioeco-nomic analysis involves linking GIS analysis with other modeling techniques.Econometric modeling and linear programming models using spatially refer-enced biophysical and socioeconomic data are examples of this type of re-search.

In the second part of the chapter, we review research that the Interna-tional Rice Research Institute is carrying out in the Mekong River Delta ofVietnam in collaboration with the Institute of Agricultural Sciences in Ho ChiMinh City. The research applies geo-informational techniques and method-ologies, and review of it enables consideration of the material covered inpart one. Here, we develop a spatial economic model of crop choice andland-use intensity to provide an analytical framework for empirical examina-tion. Empirical estimates provide insight into the roles that biophysical andsocioeconomic constraints play in explaining changes in land use and pro-ductivity, and enable exploration of the interrelation between biophysical andsocioeconomic production constraints. Random effects probit estimates showwhich factors influence farm land use. This estimator uses the panel struc-ture of the data, and provides robust estimates. Ordered probit and multino-mial logit estimates of cropping intensity and cropping pattern adopted werecarried out using single years of the survey to enable estimation of the effectof time invariant characteristics on these outcomes. Findings from theseestimates are reviewed and extensions and future applications of GIS econo-metric integration are considered.

Integration of biophysical andsocioeconomic constraintsin rainfed lowland rice farmcharacterization: techniques,issues, and ongoing IRRI researchC.M. Edmonds and S.P. Kam

442 Edmonds and Kam

Both biophysical and socioeconomic constraints influence land-use decisions andlimit the production and income of rice-farming families in Asia. Accounting for thebiophysical environment in agricultural research is essential in understanding pro-duction outcomes and the decisions of farm operators. Without proper characteriza-tion of the constraints imposed on farms by climate, topography, soils, and similarfactors, efforts to understand outcomes in order to design and disseminate new tech-nologies to bring about outcomes are likely to fail. An understanding of the socioeco-nomic environment in which farms operate is equally important in agricultural re-search, technology development, and agricultural extension. Socioeconomic constraintssuch as the availability of agricultural inputs and presence of buyers for agriculturaloutput, or noneconomic constraints such as cultural sanctions against certain castesor against women engaging in particular production activities, exert significant influ-ence over farm households. Characterizing both types of constraints and understand-ing their relation to each other is essential to the development of technologies andpolicies to increase rice production and the incomes of rice farms. Integration of tra-ditional econometric techniques with data organized in a geographic information sys-tem (GIS) offers a promising method for modeling both types of constraints.

This paper is divided into two parts. The first part provides an overview of thegeneral theme of geo-informational techniques combining biophysical and socioeco-nomic variables in the characterization of rainfed areas. We describe some of thetechniques applied in carrying out such integration, and discuss some of the problemsfaced in such research. To make these points more concrete, in part two we reviewresearch applying these methodologies that the International Rice Research Institute(IRRI) and the Institute of Agricultural Sciences (IAS) of Vietnam are carrying out inthe Mekong River Delta.

Geo-informational techniques in the characterization of rainfed environments

Geo-informational techniques, which include remote sensing, GIS, and related tech-nologies, offer a useful framework for integrating spatially referenced socioeconomicdata and biophysical information. GIS facilitates capturing the spatial dimension andspatial analysis of the effects of these factors and their interactions.

Characterizing rainfed lowland ecosystems:Which characteristics are important?In work characterizing rainfed areas, we view the objective of research to be to iden-tify biophysical constraints that limit agricultural production rather than to provide aholistic description of the environment. Agricultural scientists are accustomed to think-ing about biophysical constraints to agricultural production and to using maps ofbiophysical characteristics in designing in-field experiments or explaining variationin farm outcomes across seasons. Because the biophysical environment provides manyof the essential inputs to plant growth, the link between biophysical conditions andagricultural outcomes is relatively clear. Relative to biophysical characterization, so-cioeconomic characterization is less developed, and features a much greater range of


opinion regarding what socioeconomic characteristics are important. Agricultural sci-entists are generally less accustomed to thinking about socioeconomic constraints toagricultural production.

Because socioeconomic constraints involve human behavior, individual deci-sion making, and social norms and institutions, the task of describing the constraintsappears more daunting. The greater difficulty of socioeconomic characterization stemsnot only from the fact that it involves a description of human behavior, but also be-cause in many cases we cannot directly observe socioeconomic constraints. Instead,we must make inferences about constraints from observable social, cultural, and in-stitutional characteristics. For example, access to farmland is obviously an essentialprerequisite to farm production and land scarcity is considered to constrain the farm-ing activity of many small farms in population-dense South and Southeast Asia. Fail-ure of land markets is commonplace in developing Asian economies for several rea-sons. However, we can only observe the actual level of land the families use in theirfarming activities—when what is truly of interest is whether the household has enoughland to carry out its farming activities efficiently or is the family land constrained. Ifa household carries out its agricultural activities using only land it owns and does notengage in any land rental transactions, this could suggest several possible situations.Usually, researchers must rely on other indicators such as observed land-labor ratiosapplied on-farm combined with the observed land rental activity (or lack of activity)to make inferences about the performance of land rental markets.

The importance of location in socioeconomic characterizationSeveral instances can be cited where location and spatial relationships are of impor-tance in socioeconomic characterization and analysis. One broad spatial characteris-tic—accessibility—is described here. Other examples of instances where the locationof farm characteristics is important to consider in characterizing farms include neigh-borhood effects, production externalities that cover a geographic area, localized so-cial capital, and spatial autocorrelation (Edmonds and Kam 1999). For example, indeveloping-country agriculture, neighborhood effects can explain the adoption of newtechnologies, rural-to-urban migration patterns, and the presence of informal supportnetworks between farms. Geo-referencing farms provides a precise way of definingneighbors, whereas spatial statistics provide tools that allow flexible definitions ofneighborhood distances.

A related topic is the area of spatial econometrics. Whenever carrying out sta-tistical estimates using data for which the responses of observations vary systemati-cally depending upon their spatial location vis-à-vis other observations or centralpoints, estimation procedures that fail to take into account the systematic variationwill have inefficient parameter estimates. This systematic variation over space is re-ferred to as spatial autocorrelation. The statistical issues encountered with spatiallyautocorrelated data are similar to the more common problem of serial correlationencountered in time series data. Statistical procedures and software packages for di-agnosing spatial autocorrelation and for estimating models with spatial correlation inobservations are available (Anselin 1988).

444 Edmonds and Kam

Applications of geo-referenced biophysicaland socioeconomic data—levels of analysisGeo-information techniques provide a range of tools for creating, manipulating, andanalyzing geographically referenced data. These tools may be deployed at differentlevels of sophistication in application to socioeconomic studies.

Description and exploration of spatial patterns in variables. The simplest ap-plication of GIS with socioeconomic information is to display the geographic distri-bution of these data. The task of collecting and processing data capturing the com-plete spatial distribution of socioeconomic characteristics is cumbersome and costly,but once accomplished the display of the information on maps generated from a GISis uncomplicated and can be rewarding. Visualizing such information in a spatial andgraphical manner helps make clear the spatial pattern—if one exists—in observedcharacteristics. As an example, consider Figure 1. IRRI’s GIS facility generated thefigure from a GIS data set on rice production in northeast Thailand. It shows severalcharacteristics and spatial patterns of interest. The pie charts on the figure show therice production level (indicated by the size of the pies) by variety for 19 provinces. Itis evident that the main rice-producing provinces are in the south. Here, the predomi-nant rice variety planted is KDML105, which is a commercial, premium-qualitynonglutinous fragrant rice. Rice in the northern provinces is not only less important interms of volume produced, but in this area it is used mainly for domestic consumption(as indicated by the relatively higher proportion of glutinous varieties). The figurealso shows that KDML105 yields in the southern provinces are the lowest amongprovinces in northeast Thailand. As suggested by this example, superimposing mapsof socioeconomic outcomes and biophysical conditions can suggest research hypoth-eses and foster understanding of the relation between the two types of characteristics.

Generation of geo-referenced socioeconomic information—defining informa-tion at a common scale. A ubiquitous problem in GIS research with socioeconomicdata relates to the scale and comprehensiveness in geographic area addressed in theresearch. In particular, socioeconomic research is often asked to characterize a largearea while data available are often limited to much smaller areas. Unfortunately, mostsocioeconomic studies cover narrow geographic areas because of the difficulty ofcarrying out detailed socioeconomic surveys over large areas as the result of financialand logistical constraints. Recent advances in modern, satellite-based remote-sensingand global-positioning system (GPS) technology promise to partly alleviate short-comings in the availability of geo-referenced data. Land use and land cover, distribu-tion of human settlements, and location of infrastructure (roads, public facilities, elec-tricity, irrigation canals) are examples of the types of socioeconomic information thatare discernible using remote-sensing data. GPS technology is useful for geo-position-ing survey points, such as village and farm locations, and even in tracing field bound-aries.

The second problem in integrated GIS and socioeconomic research is that therequired data are often not available at the proper level of resolution. Commonly,aggregate data are available whereas more detailed information for small areas orindividual economic agents is required. Furthermore, data obtained from different


Fig. 1. Rice production by variety in provinces in northeast Thailand (1995-96). Source: SocialSciences Division (GIS), IRRI. 1 ha = 6.25 rai.

sources are often defined at different spatial resolutions, and need to be reconciledand integrated into a consistent data analysis framework. This requires changing thespatial reference scale of the data. There are three principal types of scale transforma-tion of data (Edmonds and Kam 1999): (1) aggregation—combining units of obser-vation to a higher scale, (2) reaggregation—changing the boundaries of observationwhile maintaining the same spatial scale or resolution, and (3) disaggregation—mov-ing data from a higher scale to a lower scale. Invariably, analyses based on trans-formed data have inherent biases and errors. Deichmann (1999) reviews three prob-lems related to aggregation bias: the modifiable areal unit problem, an error-in-vari-ables statistical problem, and the ecological fallacy problem.

1995-96 Rice production (t)TvRd6Rd15Rd21Rd23Rd-npsRd-psSb60Sb90Kdml105Ps60-1,2lv-npslv-psBm

1995-96 KDML105 yield (kg rai–1)

<200 kg/rai201–250251–300301–350351–400>400 kg rai–1

60 0 60 km

446 Edmonds and Kam

A noteworthy exception to the generalization that socioeconomic data are avail-able for only restricted geographic areas is the information collected in national cen-suses. In a national census, socioeconomic information is collected for all regions ofthe country. Unfortunately, although census data provide geographically completesocioeconomic information, few variables are collected and these generally providean incomplete picture of the socioeconomic characteristics and economic outcomesof individuals taking part in the census. Data collected in censuses generally coverlittle more than the basic demographic characteristics of families and limited infor-mation about living conditions at their place of residence. The development of tech-niques to use detailed information obtained from sample surveys in combination withextensive census data for generating data sets with more comprehensive geographiccoverage to generate poverty maps is an active area of research (Hectschel et al 1997).Development of a similar approach using data collected from remote sensing to ex-trapolate results from detailed farm surveys to broader areas is under way at IRRI.

Modeling spatial processes—integrating GIS with other analytical methods.Moving beyond description, data generation, and univariate analysis, GIS integrationwith other quantitative analytical methods used outside geography seeks to explainthe processes underlying the observed spatial distribution of variables of interest byincorporating the spatial distribution of farm characteristics in modeling farm out-comes. One means of achieving this is to link GIS with standard econometric analysisof farm survey data, that is, to use outputs from GIS as inputs into econometric mod-els, or to map the outputs of econometric analysis and use them for further spatialcharacterization. We turn our attention to research in Vietnam that applies this ap-proach in the section “Land-use dynamics in the Mekong River Delta: an illustrationof integration.”

Accessibility and transportation costsAccessibility to markets and market intelligence concerning the prices of agriculturalinputs and outputs influences the agricultural production system pursued by a farmand its outcomes. One direct influence of accessibility is on the ease, speed, and costof transportation of agricultural inputs and farm outputs. These factors influence farm-ers’ decisions regarding (i) how much of a particular agricultural commodity to pro-duce or purchase, and (ii) the type(s) of crop(s) to produce.

Influence of transportation costs on farm household decisions on production.Discussion of a simple supply and demand model displayed in Figure 2 illustratescase (i) above. The convergence of supply and demand at the economy-wide leveldefines the market-clearing price for a good (Pm). Each farm can be thought to haveinternal household supply and demand functions for goods, which depend upon thefarming resources they have available, their production efficiency, and their consump-tion needs. As the price of a good increases, the household will demand less of thegood from the market and be willing to produce more of the good for market sale. Theintersection of a household’s internal supply and demand schedules defines ahousehold’s “shadow price” for the good (Fig. 2, point c).


In the absence of transportation or other costs associated with conducting busi-ness in the market (as either a buyer or seller of a good), whether a household pur-chases a good or sells a good is defined by the relationship between the internal-household equilibrium price and the market price. A household whose shadow priceis below the market price will produce enough of the good to meet its demand as wellas provide for surplus production for sale on the market. Households for which thehousehold shadow price lies above the market price find it better to produce a portionof the good on-farm and to obtain the rest of the good by buying it at the market. Thepresence of transportation costs in bringing goods to and from the market changesthis situation. Transport costs create a “price band” around the market price. Theupper bound of the band (Pp) gives the purchase price of the good when the cost oftransporting the good from the market to the farm is added. The lower bound of theband (Ps) gives the net price of sale of the good by reducing the market price by thecost of transporting the good from farm gate to market. In terms of the effect on farmbehavior, it is unimportant whether the transport costs are paid directly by the farmersor indirectly through the purchase price offered by intermediaries purchasing the goodat the farm gate.

The higher the transport cost, the wider the price band. The bands have theeffect of lessening the amount of good bought or sold by a household. Householdsselling the good will produce and sell less of it in response to the price band (the pricedifference between Ps and Pm). Similarly, households that are purchasers of the goodwill produce more of the good themselves, purchase less of the good from the market,and consume less of the good overall. When a household’s shadow price lies withinthe price band, it will produce enough of the good to meet its demand but will notengage in market transactions for the good. Transportation costs, a spatially definedfarm characteristic, thus explain the pattern of subsistence and commercial farm pro-

Fig. 2. Supply and demand of agricultural out-put with transportation costs. (Adapted fromde Janvry et al 1991.)

Farm-gate price of agricultural good

Output purchased/sold by farm

Internal farm demand for good

Internal farm supply of good

c

Pp

Pm

Ps

Shadow price ofoutput to farm

Qs = Qd

448 Edmonds and Kam

duction in a rural area. The predictions of this simple framework are consistent withthe higher incidence of subsistence agriculture in remote areas.

Calculation and interpretation of accessibility indicators. The conventional wayof estimating transportation time or costs is from questioning of farmers or key infor-mants. This approach often faces problems of subjectivity. Alternative ways of com-puting quantitative accessibility indicators using less subjective means for estimatingtravel distance or travel cost/effort are desirable, and provide a means of verifying thedata solicited from questionnaires. Because accessibility is largely determined bygeography, that is, the availability, density, and quality of infrastructure (i.e., net-works of roads, canals), and by the nature of the terrain, geographic information sys-tems lend themselves naturally to the computation of accessibility indicators and gen-eration of accessibility maps.

Several indicators of accessibility may be computed, and they fall into twobroad categories:

1. Accessibility from the supply perspective, for example, a service area fromthe point of view of a facility, such as the serviceable area of a tube welldrawing water from an underground aquifer.

2. Accessibility from the demand perspective, for example, the ease of reach-ing or accessing services, or economic and social opportunities for the user(household, farm, and village), the travel distance or time to the nearest fa-cility or a selection of facilities, or how many markets are within a giventravel time or travel effort.

The ensuing discussion focuses on accessibility from the demand perspective,with particular emphasis on the issue of physical accessibility as a measure of thedegree of market integration and its influence on the economics of agricultural pro-duction. Accessibility is also a useful measure of the degree of integration of a givenlocation with respect to social opportunities, such as educational, health, and otherpublic services, or, conversely, a measure of isolation and deprivation.

Three indicators are available to measure the physical accessibility of rural farmsto markets (UN Statistics Division 1997): (1) the equity index, which is the distanceor travel time/cost to the nearest markets; (2) the covering index, which is the numberof markets that are accessible within a given travel time/cost; and (3) the averagetravel time/cost index, which is the average time taken or cost/effort required to get toa number of the target market(s). A fourth accessibility indicator, the potential acces-sibility index, is a more general measure of the degree of interaction or integration ofthe location of interest with target locations, and is based on a distance- and size-weighted sum of the distances to target locations. For example, one can compute anindex by summing the population of towns in the vicinity of a village, weighing eachtown’s population by an inverse measure of its distance from the village. Such anindex would indicate the extent of access of the village to contacts and social oppor-tunities, and to informal information networks and market intelligence.

All indices require reliable estimates of travel distances (dij) along availabletravel routes between an origin location i and a target location j. These distances canbe computed by using the network analysis capability of vector-based GIS by sum-


ming the distance of all line segments constituting the route between the origin andtarget locations. This requires an infrastructure map that is properly formatted as atopologically consistent network connecting the points of origin (farms/villages) withthe target or destination points (markets, schools, etc.). In remote areas, not all loca-tions are served by roads, or sometimes details of tracks are not provided in small-scale road maps. Auxiliary information such as topography, hydrography, and landcover can be used to model the “least-cost” pathway access from a remote location toexisting mapped roads. The least-cost pathways identified in a GIS can also be usedto supplement sparse road networks and provide travel distance data for better esti-mation of accessibility in areas with poorly developed transportation infrastructures.

Where the quality of transport infrastructure varies across the network and en-ables different ease of transit across it (e.g., dirt roads must be traveled at slowerspeeds than well-maintained paved roads, so that the dirt roads have a higher “imped-ance”), accessibility measures based solely on travel distances are poor indicators ofaccessibility. One way to incorporate differences in the ease of transport along differ-ent segments of the network is to “tag” the travel speed or travel times to these seg-ments, and compute the total time taken to traverse along the network from the originto the target location. The term “travel cost” is used to refer to the distance, time, orfinancial cost to get from an origin to a target location.

The ease of transporting goods from the farm gate to the market should reflectthe overall ease of transport between the farm and all the available local markets,rather than the distance or time to transport goods to the market that happens to beused by the farm at one time. Accessibility indicators calculated using several refer-ence markets are preferable because they reflect the exogenous availability of mar-kets in a particular area rather than the endogenous outcome of a farm’s choice to buyor sell goods/factors at a particular market. GIS-based analysis allows for easy com-putation of quantitative indicators that reflect overall accessibility from the farm (orvillage) to multiple markets. The average travel cost index is an example of this typeof indicator. The average travel cost index Di for an origin location i is the mean travelcost between the origin location (e.g., village) and a number (J) of target locations(e.g., markets):

where Di = average distance (in km) between farm i and the J target market(s) and dij= length/distance of the line segment k (in km) between the village and market j.

Closely related to the effect transport costs play in determining the level offarm production and the extent of subsistence orientation of farms, transport costsalso determine what crops farms cultivate and the intensity of land use on the farm.For example, farms in remote areas may be prevented from economically cultivatingfresh produce by the cost and speed required to transport output to the market. Due tothe ease of storing rice, the commodity is often particularly favored in remote areas.Agricultural marketing channels in remote areas often focus on rice. As a result, even

Di = Σ dij /JJ

j = 1

450 Edmonds and Kam

farms that have suitable conditions for growing higher value crops may nonethelesscultivate rice because of transport concerns. We return to this hypothesis shortly, whenwe review an accessibility analysis for road and river transport systems in the MekongRiver Delta.

Land-use dynamics in the Mekong River Delta: an illustration of integration

The increase in rice production in Vietnam during the 1990s represents a successstory in Asian agricultural development. Increases in rice production in the MekongRiver Delta, which supplies about half of Vietnam’s total rice production, averagedabout 6.3% per year during the nineties, according to government statistics (Govern-ment Statistical Office 1998). These increases took the country from having a largedeficit between rice demand and supply to becoming the third largest rice exporterworldwide. This expansion contributed to Vietnam’s high rate of GNP growth byproviding urban areas with cheap food and generating foreign exchange earnings.

Although the national statistics on rice production in Vietnam are widely known,there have been few studies of the farm-level changes in rice production techniquesand land-use changes that have led to production increases. This research providesinsight into the farm-level changes in agricultural production that, when aggregated,caused the production increases. We make use of previously collected longitudinalfarm-household survey data and existing biophysical characterization of the MekongRiver Delta. Farm-level changes in rice output with production, land use, and supplyestimates are examined by using data from longitudinal household survey data thatsolicited farm production information for each year between 1994 and 1997. Thesurvey covered about 150 farms from 8 villages in the Mekong River Delta and 2villages from river basin areas in Dong Nai Province. Sampled villages represent arange of agroecological and production situations. Because of nonreporting of somevillages, and to a lesser extent farm attrition from the survey, the sample size variesover time. The data were collected by the Institute of Agricultural Sciences of Viet-nam.1

Because of developments in water management infrastructure in the area, itoffers an ideal context in which to examine the effects of changes in water availabilityon agricultural production at the farm level. Different villages taking part in the lon-gitudinal survey can be taken to represent different stages of development betweenrainfed rice agriculture and irrigated farming. We can trace farm-level changes ac-companying the transition from a rainfed to irrigated agricultural rice system usingthe data. Another major development in the study area during the 1990s was the deep-ening and geographic extension of market reforms begun in 1989. These changesmake the area and time period ideal for economic study.

1The data were originally collected for a research project involving the Institute of Agricultural Sciences ofVietnam and the Faculty of Agricultural Sciences of Gembloux, University Mons-Hainaut, Belgium.


Description of the study areaWe completed an extensive descriptive analysis of the Mekong River Delta usingavailable survey, secondary-source, and GIS data. In this review, we considered char-acteristics such as the following biophysical variables: location and accessibility ofrice producers and markets, soils, rainfall and temperatures, and the seasonal flood-ing situation and problem of saltwater intrusion on farmland. We relied on GIS datacompiled by IRRI and its collaborators for most of this analysis. We also examinedsurvey and secondary price data to consider changes in agricultural policies, real pricesof rice inputs and outputs, and market development in the Mekong River Delta in the1990s, and to characterize the demographic characteristics and resource endowmentsof surveyed farms. Extensive work had previously been carried out using the surveydata to describe the evolution of rice agriculture, and in particular the evolution ofcosts and revenues accruing to rice farmers in the region during the 1990s (IAS 1997,1998). Here we report only the small portion of this work that relates directly to theGIS techniques explained earlier. Table 1 reports selected descriptive statistics fromthe data set used in the study.

Figure 3 superimposes land use reported by farms in the eight surveyed villageson a land-use map for the Mekong Delta (circa 1996). The figure helps to highlightthe benefits of integrating GIS with farm survey data. The base map provides a com-plete characterization of land use, while information from the farm-level survey addsa time dimension and provides detailed data on farm resources and activities. Thefigure shows the high level of correspondence between land use captured from re-mote sensing presented on the map and that reported by farms completing the longi-tudinal survey.

We generated rainfall indicators for each of the villages from which householdswere selected for the longitudinal survey using weekly rainfall data from 24 weatherstations in the Mekong River Delta. In this initial analysis, we worked with aggregateannual rainfall as a measure of fresh water available to nonirrigated rice plots. Theamount of rainfall is important not only as a water source for rainfed crops, but alsobecause of its influence on flood levels and saltwater intrusion in the dry season.Across the Delta, rainfall is heaviest in the far southwest coast of the peninsula andtends to decline as one goes from the Ca Mau area to the northeast (toward Ho ChiMinh City). The highest total annual rainfall in 1996 (a heavy rainfall year) was about3,100 mm, whereas the lowest rainfall reported that year was about 1,650 mm. Weused linear spatial interpolation to generate rainfall measures for the eight localitieswhere the survey was conducted. The technique takes the weighted average of re-ported rainfall at all the weather stations surrounding the surveyed villages, with theweight assigned to each weather station being inversely proportional to the distancebetween the station and the selected village.

Our description of the study area included an accessibility analysis of the eightsurveyed villages in the Delta. Figure 4 shows the main transportation routes in thearea. The principal routes used by the farms interviewed for the longitudinal surveyare shown in green (surface-water routes) and black (road and ferry routes). Theaccessibility indicators calculated for the surveyed farms in the Mekong River Delta

452 Edmonds and Kam

Tabl

e 1

. D

escr

ipti

ve s

tati

stic

s fr

om s

urve

y, s

econ

dary

, an

d G

IS d

ata.

1994 (n

= 8

9)

1995 (n

= 1

49)

1996 (n

= 1

22)

1997 (n

= 1

05)

All y

ears

Varia

ble

Uni

tM

ean

S.D

.M

ean

S.D

.M

ean

S.D

.M

ean

S.D

.M

in.

Max

.

Num

ber

of r

epor

ting

villa

ges

6 o

f 10

10 o

f 10

8 o

f 10

9 o

f 10

Hou

seho

ld d

emog

raph

ic c

hara

cter

istic

sYe

ars

sinc

e fa

mily

set

tled

y4

2.3

19

.21

85

in a

reaa

Age

of h

ead

of h

ouse

hold

ay

53

.01

5.4

21

85

Mos

t ed

ucat

ed h

ouse

hold

0/1

dum

my

0.3

29

–0

1m

embe

r (p

rimar

y)a

Mos

t ed

ucat

ed m

embe

r0/1

dum

my

0.5

77

–0

1(s

econ

dary

)aM

ost

educ

ated

mem

ber

0/1

dum

my

0.0

40

–0

1(p

osts

econ

dary

)aTo

tal p

erso

ns r

esid

ing

inIn

divi

dual

s4

.71.5

25.8

1.7

95.8

1.7

15.6

1.6

62

13

hous

ehol

dLa

nd-la

bor

ratio

in h

ouse

hold

ha–1

wor

ker–

10

.36

0.3

10

.40

0.5

20

.34

0.4

50

.40

0.4

80

3.9

6

Land

hold

ing

and

biop

hysi

cal c

hara

cter

istic

sTo

tal l

and

owne

d by

fam

ilyha

0.9

10

.63

1.2

20

.79

1.1

50

.70

1.0

70

.64

0.1

34

Farm

ing

plot

s cu

ltiva

ted

by f

amily

no.

1.3

0.6

91

.50

.79

1.1

0.4

51

.00

.10

15

Qua

lity-

adju

sted

land

hold

ing

size

Qua

lity-

adj.

ha1

.14

1.1

11

.47

1.9

21

.30

1.8

11

.34

1.9

10

.04

17

.84

Allu

vial

soi

lb0/1

dum

my

0.5

1–

01

Med

ium

-slig

htly

aci

d su

lfate

soi

lb0/1

dum

my

0.1

0–

01

Sal

ine

soils

with

dry

-sea

son

0/1

dum

my

0.2

1–

01

saltw

ater

b

Ric

e pr

oduc

tion,

mar

ketin

g, a

nd la

nd u

sePa

ddy

yiel

d du

ring

win

ter-s

prin

gkg

ha–

13

,84

11

,38

25

,28

81

,49

05

,67

01

,07

35

,02

31

,37

71

,05

39

,00

0Ar

ea c

ultiv

ated

to

rice

autu

mn-

ha0

.88

0.5

51

.00

0.6

10

.92

0.6

20

.79

0.4

10

.12

53

.5w

inte

rTo

tal y

early

ric

e pr

oduc

tion

in1

,00

0 M

T–

–1

0,5

29

5,4

30

13

,79

23

,61

51

1,5

39

5,9

39

01

8,0

32

prov

ince

Ric

e-cr

oppi

ng in

tens

ityno

.1

.80

.79

1.9

0.6

91

.90

.76

2.1

0.5

80

3C

ultiv

ated

non

rice/

nonr

ow c

rop

0/1

dum

my

0.1

2–

0.5

5–

0.5

9–

0.5

6–

01

Padd

y so

ld b

y fa

rm d

urin

g ye

arkg

––

5,4

59

5,9

01

5,5

41

5,4

86

5,4

05

5,1

06

03

2,0

60

cont

inue

d on

nex

t pa

ge


Tabl

e 1 c

onti

nued

.

19

94 (n

= 8

9)

1995 (n

= 1

49)

1996 (n

= 1

22)

1997 (n

= 1

05)

All y

ears

Varia

ble

Uni

tM

ean

S.D

.M

ean

S.D

.M

ean

S.D

.M

ean

S.D

.M

in.

Max

.

Av s

ale

pric

e of

pad

dy d

urin

g’9

7 U

S$ k

g–1

0.1

30

.03

0.1

70

.04

0.1

40

.02

0.1

50

.02

0.0

90

.30

year

c

Av lo

cal m

arke

t pa

ddy

pric

e’9

7 U

S$ k

g–1

0.1

40

.01

0.1

80

.02

0.1

50

.01

0.1

40

.01

0.1

20

.20

durin

g ye

ar

Agric

ultu

ral t

echn

olog

y, p

ract

ices

, an

d in

puts

Trad

ition

al n

ongl

utin

ous

rice

0/1

dum

my

0.1

8–

0.1

1–

0.1

1–

0.0

7–

01

Mod

ern

shor

t-dur

atio

n0

/1 d

umm

y0

.20

–0

.35

–0

.37

–0

.43

–0

1no

nglu

tinou

s ric

eM

oder

n m

ediu

m-lo

ng-d

urat

ion

0/1

dum

my

0.5

2–

0.4

0–

0.3

4–

0.2

4–

01

nong

lutin

ous

rice

Ure

a ha

–1 y

ear–

1 a

vkg

ha–

11

60

87

16

07

41

69

71

14

97

30

53

3Pr

ice

of u

rea

(wei

ghte

d ye

arly

av)

’97 U

S$ k

g–1

0.2

44

0.0

28

0.2

36

0.0

22

0.2

48

0.0

16

0.1

96

0.0

23

0.1

23

0.3

37

Loca

l mar

ket

pric

e of

ure

a’9

7 U

S$ k

g–1

0.3

55

0.0

09

0.3

02

0.0

05

0.2

32

0.0

14

0.1

87

0.0

05

0.1

75

0.3

69

(yea

rly a

v)N

o. m

echa

nize

d tr

acto

rs u

seda

0/1

dum

my

0.2

8–

01

Whe

ther

hom

este

ad o

wns

0/1

dum

my

0.2

6–

01

dryi

ng c

ourt

a

Wat

er m

anag

emen

t in

fras

truc

ture

Land

leve

ling

carr

ied

out

on-fa

rm0

/1 d

umm

y0

.38

–0.2

8–

0.2

9–

0.3

8–

01

Dik

e co

nstr

uctio

n on

far

m0

/1 d

umm

y0

.48

–0.3

6–

0.3

4–

0.3

9–

01

Inte

rpol

ated

ann

ual r

ainf

all a

tm

m1

,25

16

61

,61

62

16

1,8

63

10

71

,51

31

45

1,1

74

2,0

76

villa

geFl

oodi

ng 0

.5–1

m la

stin

g 3 m

ob0/1

dum

my

0.1

2–

01

Bra

ckis

h (>

4 g

L–1

) w

ater

>6 m

ob0/1

dum

my

0.1

3–

01

Rai

nfed

far

m (

no ir

rigat

ion)

b0/1

dum

my

0.1

1–

01

Lim

ited

irrig

atio

n av

aila

bleb

0/1

dum

my

0.1

0–

01

Rel

iabl

e irrig

atio

n on

-farm

b0/1

dum

my

0.6

1–

01

Mar

ket

acce

ssib

ility

and

tra

vel d

ista

nces

Dis

tanc

e to

nea

rest

loca

l mar

ketb

km19.8

4.6

12

28

Acce

ssib

ility

inde

x—di

stan

ce t

okm

19.5

4.7

11

28

near

est

mar

ketb

Acce

ssib

lity

inde

x—tim

e to

all

min

13

0.2

70

.25

12

55

loca

l mar

kets

b

a Num

bers

repo

rted

com

e fr

om (1

) a b

asel

ine

surv

ey th

at d

id n

ot in

clud

e al

l hou

seho

lds

late

r int

ervi

ewed

for t

he lo

ngitu

dina

l sur

vey,

and

(2) i

nter

pola

tion

or o

verla

y of

val

ues

gene

rate

dfr

om G

IS c

over

ages

. N

umbe

r of

obs

erva

tions

for

par

ticul

ar v

aria

bles

can

var

y fr

om g

ener

al s

ampl

e si

zes

repo

rted

. b S

oil,

wat

er a

vaila

bilit

y, a

nd a

cces

sibi

lity

mea

sure

s w

ere

deriv

edfr

om G

IS c

over

ages

ava

ilabl

e in

Mek

ong

Del

ta p

rovi

nces

onl

y. c

Cal

cula

ted

as t

he w

eigh

ted

aver

age

(by

quan

tity

of r

ice

sold

) sa

le p

rice

of r

ice

repo

rted

by

surv

eyed

far

ms.

454 Edmonds and Kam

Fig. 3. Land-use map in the Mekong River Delta in the 1990s. Land-use map and land usereported by surveyed farms. Source: V.Q. Minh, Soil Science Department, College of Agricul-ture, Can Tho University, Can Tho, Vietnam.


Fig. 4. Main transport routes between surveyed villages and main markets. Source: IAS-IRRIstudy.

Cambodia

456 Edmonds and Kam

apply the measures based on travel distance and time between single markets (thenearest local market to the farm and Ho Chi Minh City) and average distances/timesfor transport between the farm and all surrounding markets. The accessibility analy-sis and calculation of accessibility indicators were already discussed, so we merelyreport the results of the analysis in Table 2.

Analytical model of land useTo explore the relationship between biophysical and socioeconomic characteristics,to derive hypotheses, and to form an estimation equation that can be tested usingavailable data, we develop a spatial land-use model following standard microeconomicanalysis. Spatial economic models assert the importance of the spatial location ofagents to market centers, economic infrastructure, and to one another in determiningthe economic activities pursued by the agents. They offer a good analytical frame-work for considering the effects of biophysical characteristics of a parcel of land andsocioeconomic characteristics of the farm operator(s) on land use.

Building upon the insights of von Thünen (1826) and more recent work ofChomitz and Gray (1995), our model examines the effect of travel distances betweenfarms and markets on cropping patterns and land-use intensity of farms. It is beyondthe scope of this chapter to review the model in detail, but we outline its structure anddevelopment. We begin by assuming that land will be used for the activity that gener-ates the highest rent given the physical characteristics of the plot (local climate, basisof land tenure, labor available for farming) and that farm-gate input and output pricesdepend upon the cost of transport.2 We define a revenue function for each alternativeuse of the plot. In notation, we define

Rik = PikQik(Pik,Cik,Zi) – CikXik(Pik,Cik) + uik (1)

where Rik gives the rent on plot/point i in use k, Pik is the price of output/crop k at plot/point i (farm-gate price of k), Cik is a vector of prices of inputs needed for productionof crop k at plot/point i, Zi is a vector of fixed characteristics of the plot that influencethe land’s production efficiency in use k, Xik is the optimal input level for productionof crop k per unit of land at point i, Qik is the potential output of crop k at plot/point i(potential production), and uik is a random disturbance term.

2Several other assumptions also underpin the model. The model assumes that land use is reversible—thatonce land has been applied to a particular crop it is possible to alter the use of the land in the future. The basisof land tenure affects only the profit the farm operator obtains from cultivation of each crop. This assumptioncan be captured in the model by defining the farm-gate price for each crop to be net of any rental costs that thefarm operator must pay. The model also requires strong assumptions concerning the farm operator’s expecta-tions about future prices of goods that can be produced on the land and of inputs required to produce thosegoods.


Tabl

e 2

. A

cces

sibi

lity

indi

cato

rs for

sur

veye

d vi

llage

s—di

stan

ce a

nd t

rave

l tim

es.

To n

eare

st lo

cal m

arke

tTo

Ho

Chi

Min

h C

ityTo

all

loca

l mar

kets

Tim

e of

Tim

e of

Villa

getr

avel

(m

in)

trav

el (

min

)Ac

cess

ibili

ty in

dex

(min

)N

ame

ofD

ista

nce

Dis

tanc

eD

eman

d/su

pply

mar

kets

mar

ket

(km

) T

ime

Tim

e(k

m)

Tim

eTi

me

Tim

eTi

me

Dis

tanc

e1

21

21

2(k

m)

Duo

ng X

uan

Hoi

Tan

An1

2.4

30

30

57

.59

77

5Ta

n An

, H

o C

hi M

inh

City

, M

y Th

o6

55

12

9.8

Thua

n M

yTa

n An

60.6

165

120

Tan

An, H

o C

hi M

inh

City

, M

y Th

o1

26

82

37

.6vi

a ro

ad2

2.0

90

60

via

cana

l2

9.6

11

889

Than

h Q

uoi

Thot

Not

22

.73

02

71

94

.02

98

21

6Th

ot N

ot,

Than

h Th

ang,

Lap

Vo,

53

47

37

.0R

ach

Gia

, C

an T

ho,

Sa

Dec

Vinh

My

AB

ac L

ieu

Bac

Lie

u, C

a M

au, S

oc T

rang

25

52

46

44

.0vi

a ro

ad1

7.2

30

30

29

4.6

45

53

08

via

cana

l1

6.9

18

01

80

29

4.3

60

54

58

Hoa

Kha

n Ta

yD

uc H

oa T

own

17

.93

627

46.8

94

70

Duc

Hoa

Tow

n, T

an A

n, H

o C

hi108

64

41.2

Min

h C

ityD

uc L

ap H

aD

uc H

oa T

own

16.6

34

25

Duc

Hoa

Tow

n, H

o C

hi M

inh

City

51

36

29

.0D

uc L

ap H

a to

39

.79

26

9D

uc H

oaTo

wn

to H

CM

Duc

Lap

Ha

to C

u4

1.4

67

46

Chi

to

HC

MTh

anh

Xuan

Cai

Ran

g To

wn

21

.28

55

11

96

.03

78

24

0C

ai R

ang,

Can

Tho

, Ph

ung

Hie

p,1

81

10

94

7.9

(brid

ge)

Ca

Mau

Hoa

An

Phun

g H

iep

28

.51

14

69

206.6

420

265

Can

Tho

, Ph

ung

Hie

p, R

ach

Gia

,1

88

12

46

9.2

Vi T

hanh

458 Edmonds and Kam

The prices of inputs and outputs in the revenue equation depend upon the dis-tance of the land from the market. Prices of inputs increase at an increasing rate asfarms move farther away from markets. Similarly, the prices farms can obtain foroutput they sell are assumed to decrease at an increasing rate as the farm is moredistant from markets. There are two components to this distance—the distance be-tween farm plots and the homestead, and the distance between the homestead and thereference market(s). In the Mekong River Delta, transport of paddy and rice inputs islargely by boat. Settlement patterns in the area usually focus on canals—with home-steads usually bordering a canal or river.

We specify a functional relationship between the level of input applied to farm-ing and the amount of output produced by the farm. The level of output produceddepends upon input levels, agroclimatic conditions, and other fixed land characteris-tics. Using the production function and the expressions for net revenue associatedwith cultivation of each crop, we can derive relationships between the factors speci-fied as determining net revenue and production, and the demand for inputs by thefarm. The demand for inputs for crop k cultivated at location i is a function of the costof the inputs, the farm-gate price of the output, the characteristics of the plot, and theefficiency of production of crop k on the plot. Next, using the expressions for inputdemand, production, and the effect of travel distances on revenues, we define anexpression for the net returns associated with cultivation of crop k on parcel i, whichincorporates the effects of travel cost and the production technology of the farm. Twotravel distances are considered in the model. Di is the distance between the homesteadand the farming plot or plots operated by the family, whereas Ti is the average dis-tance between the homestead and the input/output market(s) accessible to the farm.Both distances are relevant in the model since various inputs used in farming (e.g.,labor, fertilizer, seed, etc.) and the outputs produced are transported between home-steads, farm plots, and markets over the course of a production season. The resultingexpression establishes the hypothesis that the likelihood that a plot will be applied tocultivation of a particular crop and the intensity of use will fall as the distance be-tween the plot and the output/input market increases. At the extreme, very distantplots will not be cultivated, whereas plots closest to markets would be expected to beused for intensive commercial farming.

We form an expression for net revenue from cultivation of crop k on plot iamenable to estimation from earlier equations:

ln(Rik) = a0k + a1kln(Di) + a2kln(Ti) + a3kln(z1i) + a4kln(z2i)+ … + aNkln(zLi) + uik (2)

If we add technical assumptions concerning the distribution of error terms (uik) andthe correlation of errors, the probability of any crop k being cultivated on plot i isdistributed according to the multinomial logit distribution. This provides the basis forusing the multinomial logit model in empirical tests of the model. If we can rank thealternative land uses—as is possible when the sample is limited to farms cultivating


rice and the model is applied to explain rice-cropping intensity—the model can bemodified to take the form of an ordered logit model.

Under the model, the coefficients on distances (Di and Ti) are expected to benegative, whereas those on productivity-enhancing land characteristics (sik) are ex-pected to be positive. The magnitude of the estimation coefficients will depend uponper unit costs of transportation of different crops and the relevance of a particular landcharacteristic to the production of a particular crop. Whether the crop being culti-vated on the plot is destined for commercial or subsistence use will also affect theinfluence of distance on the likelihood that a particular crop is produced. Subsistencecrop production is less influenced by distance.

Why develop an analytical model? The model provides a framework for em-pirical estimates of the determinants of land use in the Mekong Delta, but is notintended to provide a complete description of reality. Actual decision making of farmsregarding land use is extremely complex—much more complex than the simple staticmodel just outlined. For example, farmers’ land-use decisions incorporate consider-ations of the dynamic effects of using land for a particular purpose in this period on itsproductivity in future years under alternative uses. The role of agricultural income inthe broader income of the household also influences land use, as do farmers’ concernsabout the risk and expected revenue from alternative land uses. However, given theavailable data with their measurement error, missing data problems, and relativelysmall sample size, it is necessary to simplify the number of variables included in theanalysis and the number of issues the analysis considers. The model provides thisfocus. To facilitate interpretation of estimation results, modeling the relationshipsbetween variables thought to be important is essential. The model establishes hypoth-eses that can be tested in the estimates and be used to evaluate whether the structuredeveloped by the model is consistent with available data. This parameterization of therelationship between variables and of the estimation comes at the cost of restrictingthe number of variables and possible relationships between variables that can be con-sidered in the analysis.

Empirical model and hypothesis testsThe analytical model just reviewed provides the basic framework we apply in analyz-ing farm survey data, and establishes the multinomial logit and ordered probit estima-tors as appropriate for estimating land use and cropping intensity. The form of theestimation equation is given by equation (2). The key variables of interest in esti-mates are the terms detailing the distances between the homestead and the farmingplots, and the distances between the homestead and markets accessible to the farm.The exogenous or predetermined z variables in equation (2) are other household orfarm characteristics expected to influence household land-use decisions, and includecharacteristics of the biophysical environment where farms are located, family char-acteristics, and variables capturing market conditions in surveyed villages. Standardmicroeconomic analysis of production and supply also guides the selection of vari-ables and our expectations regarding their signs in estimations, but we do not review

460 Edmonds and Kam

these, and production and supply function estimation results are not reported. Theparticular set of right-hand-side variables employed in our empirical estimates variesdepending upon the relevance of variables to the left-hand-side variable we are seek-ing to explain. In some cases, we had to reduce the number of right-hand-side vari-ables included in the estimation equation in order for the estimator to find a maxi-mum. This resulted from difficulties caused by missing data and the relatively smallsample size available (vis-à-vis the estimation procedures applied).

Empirical estimates use both cross-sectional and panel data-based estimationprocedures. Panel data estimation procedures provide more robust estimates becausethey can control for the effect of variables that cannot be observed but are known tohave a strong effect on estimated outcomes (e.g., household preferences and motiva-tion). Because panel data estimators make use of the full panel of data (rather thansingle years of the survey), they can measure more precisely the effect of changes inexplanatory variables that can explain change in the variable of interest. Despite theseadvantages, we make use of cross-sectional data-based estimators in our empiricalanalysis for two principal reasons. First, many of the right-hand-side variables ofinterest—including our accessibility indicators and most of the variables capturingaspects of the biophysical environment in which farms operate—could be observedonly a single time during the years of the survey. Panel data estimators cannot accom-modate the use of time-invariant right-hand-side variables in estimation equations.Second, for technical reasons, certain estimation procedures (e.g., the multinomiallogit estimator) have not been developed for panel data. This makes it necessary forus to define our land-use categories so that only two alternative land uses are avail-able to apply a panel probit estimator, and to estimate land use defined over more thantwo categories using single years of the survey with cross-sectional estimators.

In our estimates, we categorize cropping patterns and land uses by intensityrankings (e.g., mono-cropping, double-cropping, etc.) and according to type of cropcultivated. Crops are divided into broad categories: (1) rice, (2) upland row crops(e.g., sugarcane, potato, vegetables), and (3) fruit trees or perennial fruit crops (e.g.,dragon fruit) or trees maintained by farms for wood (e.g., eucalyptus). We cannotreview all our estimation results here, but we do review panel and cross-sectionalestimates of rice-cropping intensity in Tables 3 and 4 as examples of the analysis weperformed.

We discuss the results from the panel data estimators first. Three models areestimated using the random effects probit estimation procedure: (1) whether the farmreported cultivating one or more than one crop per year, (2) farm cultivation of nonricecrops, and (3) whether the farm reported growing fruit trees or other perennial crops.We used a common set of right-hand-side (explanatory) variables in these estimates.Because of the number of parameters that need to be estimated to account for house-hold-specific error terms, we faced significant constraints in the number of right-hand-side variables that could be considered in panel estimates. An additional con-straint to the variables used in panel data-based estimators is that they be observedover time.


The variables considered are the land-labor ratio on the farm (area per full-timeequivalent family worker), the age of the head of household, a series of dummy vari-ables indicating the main rice variety cultivated (the omitted variety classes are gluti-nous and traditional nonglutinous rice varieties), and dummy variables indicatingfarm investment in dikes or land leveling. It is expected that households with lowerland-labor ratios will adopt more labor-intensive land use. It is argued that older farmoperators are more traditional and hesitant to adopt new technologies and changetheir agricultural practices. Their reluctance may also reflect their lower levels ofeducation.

Table 3. Summary of estimates of land use using the random effects probit estimator.

Cultivates Cultivates Farm cultivatesLeft-hand-side dependent variable more than nonrice fruit or other

one crop y–1 crop(s) treesEstimation coefficient 1994-97a 1994-97 1994-97 (estimated standard error) (n = 436) (n = 436) (n = 436)

Land-labor ratio on farm 1.204** –1.058* 0.056(ha per household laborer) (0.559) (0.578) (0.591)

Age of head of household 0.009* –0.050*** 0.014**(0.005) (0.014) (0.007)

Farm used short-duration 1.438*** –0.131 –0.018 modern variety seed(s) (0.343) (0.670) (0.452)Farm used medium- or long-duration –0.218 –2.221*** –0.334

modern varieties (0.317)*** (0.775) (0.364)Farm invested in land leveling 0.664 –0.116 –0.902**

or other soil improvement (0.462) (0.467) (0.450)Farm invested in construction 0.031 0.646* –0.769*

of dikes or irrigation (0.297) (0.381) (0.449)Rho 0.703*** 0.787*** 0.970

(0.153) (0.225) (0.081)Goodness of fit diagnostics:

Pseudo R2: Cragg-Uhler 0.085 0.130 0.447Maddela 0.046 0.064 0.317McFadden 0.060 0.098 0.309

Likelihood ratio (X2) test 20.653*** 28.693*** 166.549***[degrees of freedom] [1] [1] [1]

Percentage correctly predicted 0.812 0.842 0.672

Actual/predicted 0 1 total 0 1 total 0 1 total

0 0 82 82 342 17 359 107 126 2331 0 354 354 52 25 77 66 137 203

total 0 436 436 394 42 436 173 263 436

a*** = estimated coefficient is statistically significant at 99% confidence level, ** = estimated coefficient isstatistically significant at 95% confidence level, * = estimated coefficient is statistically significant at 90%confidence level.

462 Edmonds and Kam

The rice variety planted by farms clearly influences the number of rice crops itis possible for the farm to cultivate. The two dummy variables used in the reportedestimates define farms growing short-duration modern varieties and medium- or long-duration modern varieties. Rice variety choice is endogenous with the choice of crop-ping pattern, so estimates are open to endogeneity bias under the present specifica-tion. Unfortunately, neither the data nor the estimation procedures available enable

Table 4. Summary of estimates of rice-cropping intensity using the ordered probit estimator forcross-sectional data.

Left-hand-side/dependent Rice-cropping Rice-cropping Rice-cropping Rice-cropping variable intensity intensity intensity intensityEstimation coefficient in 1994a in 1995 in 1996 in 1997 (estimated standard error) (n = 60) (n = 114) (n = 114) (n = 77)

Constant 7.2873 –17.884*** 16.599*** –34.870***(5.559) (4.217) (4.678) (11.088)

Average distance between 0.0011 –0.131 –0.062 –0.025homestead and plot or plots (0.118) (0.138) (0.139) (0.185)

Average travel time to all 0.0323 –0.041*** 0.020*** –0.060***accessible local markets (0.033) (0.011) (0.007) (0.027)

Land-labor ratio on farm 0.0308 0.689 0.356 –0.100(ha per household laborer) (0.992) (0.521) (0.464) (0.834)

Years since family settled in 0.0064 –0.007 –0.003 –0.012current place of residence (0.011) (0.007) (0.007) (0.014)

Maximum educational attain- 0.0020 0.311 0.071 0.526ment of any family member (0.532) (0.296) (0.294) (0.461)

Whether farm served by good- 2.2076 2.053*** 4.266*** 4.704***quality irrigation system (1.513) (0.466) (0.843) (1.668)

Annual precipitation where –0.0091 0.013*** –0.011*** 0.025***farm is located (0.006) (0.003) (0.003) (0.008)

Mu (threshold parameter for 0.3267** 2.208*** 1.801*** 4.765**first stratum) (0.149) (0.330) (0.309) (2.182)

Goodness of fit diagnostics:Pseudo R2: Cragg-Uhler 0.432 0.633 0.521 0.768

Maddela 0.361 0.555 0.460 0.654McFadden 0.248 0.386 0.288 0.556

Likelihood ratio (X2) test 26.862*** 92.256*** 70.189*** 81.707***[degrees of freedom] [7] [7] [7] [7]

Percentage correctly predicted 0.800 0.719 0.632 0.805

Actual/ 0 1 2 total 0 1 2 total 0 1 2 total 0 1 2 totalpredicted

0 37 0 0 37 21 12 0 33 18 13 0 31 6 6 0 121 6 0 1 7 4 45 6 55 5 35 11 51 2 38 5 452 5 0 11 16 0 10 16 26 0 13 19 32 0 2 18 20

total 48 0 12 60 25 67 30 114 23 61 30 114 8 46 23 77

a*** = estimated coefficient statistically significant at 99% confidence level, ** = estimated coefficient statis-tically significant at 95% confidence level, * = estimated coefficient statistically significant at 90% confidencelevel.


estimation of a system of multinomial equations using panel data. The estimates alsoinclude farm-specific error estimates and a parameter (Rho) that provides an indica-tor of the significance of farm-specific error estimates. A statistically significant esti-mate of Rho supports the inclusion of the farm-specific error terms and use of therandom effects estimator.

The three models were each highly statistically significant. At the bottom ofTable 3, we report several measures of the overall performance of the models in ex-plaining land use. The likelihood ratio test can be interpreted as testing the null hy-pothesis that the model, as a whole, cannot explain the variation in the left-hand-sidevariable of interest. All three of the models rejected this null hypothesis at a high levelof statistical significance. The psuedo-R2 measures report nonlinear measures of theproportion of variation in the land-use variables that is explained by the overall model.Psuedo-R2 measures vary between 44.7% and 4.6% across measures and models.Lastly, Table 3 reports the share of land-use categories correctly predicted by eachmodel, and the distribution of actual versus predicted land use.

T-tests of the null hypothesis that each right-hand-side variable has no effect ona farm’s chosen land use provide the basis for testing the significance of each vari-able. These are computed using the estimated coefficients and standard errors. Therandom effects probit estimator is nonlinear, which makes it difficult to interpret esti-mation coefficients directly. We must approximate the marginal effect of a change ina right-hand-side variable on the probability that a farm chose a particular land use atthe mean values of the right-hand-side variables using an approximation algorithm.The estimates of whether the farm cultivated more than a single crop during the agri-cultural year show that the land-labor ratio, the head of household’s age, and the useof modern short-duration rice varieties each increased the likelihood that a farm mul-tiple-cropped its land. Construction of water control dikes and carrying out land lev-eling were also associated with an increased likelihood of multiple cropping, but theseeffects were not statistically significant. The estimated marginal effect of a 1% in-crease in the land-labor ratio of farms is to increase by 13.3% the likelihood that thefarm cultivated more than a single crop per year. An increase of 10 years in the age ofthe household head was associated with only a 0.1% increase in the likelihood ofmulticropping on-farm. Farm use of short-duration modern rice was associated with a15.9% increase in the likelihood that the farm engaged in multicropping. Interpreta-tion of other models reported in Table 3 follows this analysis, but we forego discus-sion because of space constraints.

The model estimates reviewed provide insight into the characteristics associ-ated with the land-use choices made by farms. We find that farm size and, more par-ticularly, the relative abundance or scarcity of family agricultural labor in relation tothe land operated by the farm play an important role in driving farm land use. Farmswith relatively abundant labor compared with their farm size are less likely to engagein multicropping. The choice of modern rice varieties and the growing duration ofvarieties chosen are closely related to the broader land-use choice of farms. Finally,investments in farm or plot improvements to water management were also clearlylinked to land-use choices. One of the benefits of dike construction, for example,

464 Edmonds and Kam

appears to be the opportunities it creates for farms to cultivate nonrice crops. In theabsence of such investments, farms appeared to adopt land-use options (i.e., fruittrees and other perennial crops) more immune to the effects of poor water manage-ment. Lastly, the significance of the parameter Rho emphasizes the importance ofunobservable farm/household characteristics on land-use options and underscores thecomplexity of the land-use choices of farms.

Our measures of market accessibility and variables characterizing biophysicalconditions in the surveyed villages were fixed over time or observed only once, so weare unable to examine the principal hypotheses of our analytical model through thepanel estimators. Accordingly, we used cross-sectional estimates of cropping patternsand rice-cropping intensity to examine our hypotheses concerning the effect of acces-sibility of land use. Rice-cropping intensity is a categorical variable for which thecategories have a natural ordering, so an ordered probit estimator is used. The esti-mates of rice-cropping intensity are significant overall in each of the four years ac-cording to the goodness of fit measures reported at the bottom of Table 4.

The key variables of interest from our analytical model are the measures of thedistance between the villages where farms are located and the average travel time toall local markets accessible to the farm and the distance between the homestead andplot or plots operated by the farm. The greater these distances, the lower the likelyrice-cropping intensity to be adopted by the farm. Estimation results provide limitedconfirmation of the model’s hypotheses. The distance between the farm plot and mar-kets had a negative and statistically significant effect on rice-cropping intensity inmodel estimates in 1995 and 1997. According to 1995 results, each 10 km of distancebetween the homestead and accessible markets causes a 4.4% reduction in the prob-ability that the farm cultivated two rather than a single rice crop, and a 6.7% reductionin the probability of cultivating three rather than two rice crops. Similar marginaleffects were estimated in 1997. Greater distances between farm homesteads and plotswere associated with a reduced probability of intensive rice cultivation by the farm—but estimated parameters were not statistically significant. According to 1995 estima-tion results, a 10-minute increase in the average travel time between the farm andplots was associated with a 14% and 21% decrease in the probability of cultivatingtwo and three crops during the year, respectively. Across all the land-use modelsestimated, the distances between the farm and markets and between the homesteadand farm plots had the effect of reducing the cropping intensity on-farm. The effect ofthe distance between the farm and markets was greater in the case of rice-croppingintensity, whereas distances between the homesteads and farming plots caused greaterreductions in the general cropping intensity (results not shown).

The study estimated other land-use, production, and rice supply functions im-plied by our land-use model or basic microeconomic theory. These included the fol-lowing: general cropping intensity, farm cropping pattern, and rice production andfarm supply of rice to the market. Together, the estimates provide a clear indication ofthe factors driving farm land-use, production, and marketing decisions. We summa-rize the major conclusions that emerge from these estimates.


The availability of low-saline irrigation water to farms had a positive and statis-tically significant effect on the intensity of land use applied by surveyed farms acrossall our land-use estimates. The magnitude of the effect of high-quality irrigation oncropping intensity was much greater than the effects of other explanatory variablesincluded in the model. Rainfall levels had mixed effects on the cropping intensity ofsurveyed farms. In years with normal to high rainfall, increased rain was associatedwith increased cropping intensity. In 1996, however, rains were particularly heavyand higher rainfall in that year was associated with significantly reduced levels ofcropping intensity among surveyed farms. The size of families’ landholdings relativeto their available labor had mixed signs across estimates and years, but generallysupported the hypothesis that the relative scarcity of land to labor leads to more inten-sive land use. Results showed that rice variety selection was clearly linked to crop-ping intensity, with the adoption of modern short-duration rice varieties playing a keyrole in enabling more intensive rice cultivation. Farm-level investments in land level-ing or dike construction were shown to increase the likelihood that farms adoptedintensive rice agriculture in the panel data-based land-use estimates. Other variablessuch as the level of education in the household, the age of the household head, or thefarming experience of the family did not have consistent statistically significant ef-fects on land use in our estimates.

Rice production and supply estimates were able to explain most of the observedvariation in the levels of rice output and marketed surplus across surveyed farms.Adjusted R2 coefficient estimates across the production and supply models rangedbetween 0.70 and 0.93, with panel data estimators performing better than estimatorsusing only single years of the survey data. Variables estimated to have positive andstatistically significant effects on the level of output included farm size, the croppingintensity pursued by farms, the amount of hired labor used in crop cultivation, and thelevel of seed application. The use of modern versus traditional varieties did not havea consistent positive effect on rice production. The principal effect of the use of mod-ern (usually short-duration) varieties appeared to be to enable farms to pursue moreintensive rice production. Other variables included in the production estimates suchas amount of fertilizer or pesticide applied on the crop had positive and statisticallysignificant effects on rice production in only a few of the production estimates. Theestimated price elasticity of supply ranged between 0.145 and 0.319 in year-aggre-gate supply estimates. The rice price had a consistent positive and statistically signifi-cant effect on rice marketed surplus, as did the quantity of urea applied to the crop.When estimates included the price of urea, occasional positive and statistically sig-nificant coefficient estimates were obtained. The increasing price of urea during theyears of the survey appears to have been dominated by a broader trend toward in-creasing urea use among surveyed farms during these years.

Simulation model for evaluation of investmentsThe implications of model estimates can be better understood by using them to for-mulate a simulation model to assess the effect of policy changes or investments ininfrastructure on land use and rice production. The results of a simulation model

466 Edmonds and Kam

derived from our empirical estimates are summarized in Tables 5 and 6. Table 5 showsthe distribution of rice mono-, double-, and triple-cropping among surveyed farmsacross the years of the survey. The actual distribution of farms in each of the fouryears is shown, along with the projected distribution (using model estimates reportedin Table 4) under alternative scenarios. One scenario involves improvements in travelnetworks between surveyed villages and local markets. The second considers the ef-fect of land transport improvements or land consolidation that brings homesteads andfarm plots closer. The third contemplates extension of water management infrastruc-ture to an additional 10% of the surveyed farms.

We use results from production function estimates, combined with the impliedchanges in the share of farms double- or triple-cropping rice, to calculate an impliedchange in aggregate rice output across farms under the different scenarios. The pro-duction estimates provide a measure of the average change in annual rice yield asso-ciated with mono-, double-, or triple-cropping of rice. The changes in total rice pro-duction from the scenarios are detailed in Table 6.

The simulation model shows the large effect of investments in irrigation exten-sion on rice production levels, and the more moderate effects obtained from improve-ments in the transportation system or land consolidation. It shows how the results ofland-use and production estimates based on the observed behavior of farms can beused to assess the likely effects of different investments on rice production levels.The integration of behavioral parameters from econometric estimates is broadly ap-plicable and can provide a needed empirical basis for larger simulation models. In-corporating the estimates obtained in this research with other linear programming orgeneral simulation models would be an important application of this research.

Conclusions

Review of the IAS-IRRI research on land use in the Mekong River Delta illustratesmany of the geo-informational techniques proposed for integrating analysis of bio-physical and socioeconomic constraints to rainfed rice production. The study com-bines farm survey and GIS data and goes beyond description in exploring the causalrelationships between the two types of factors. This study relied on existing sourcesof data that were originally collected for a cost-price accounting study or for generalcharacterization of biophysical conditions in the Mekong River Delta. Because ofthis, we encountered severe data constraints in carrying out our analysis. Despitethese limitations, the research has provided insights into the farm-level changes inland use and production systems that enabled rice production to increase in the MekongRiver Delta in the 1990s. The quantitative analytical techniques developed in thisresearch offer promise for applications in future research.


Tabl

e 5

. S

imul

atio

n of

eff

ects

of in

vest

men

ts o

n di

stribu

tion

of fa

rm r

ice-

crop

ping

inte

nsit

y.

Pred

icte

d di

strib

utio

nPr

edic

ted

dist

ribut

ion

Pred

icte

d di

strib

utio

nif

10%

incr

ease

inif

aver

age

trav

el t

ime

if di

stan

ce b

etw

een

no.

of f

arm

s se

rved

by

Ric

e-cr

oppi

ng p

atte

rnto

mar

ket

is r

educ

edho

me

and

plot

(s) is

good

-qua

lity

irrig

atio

nAc

tual

dis

trib

utio

naby

10 m

inut

esre

duce

d by

1 k

msy

stem

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

Mon

o-cr

oppi

ng3

73

33

11

23

62

12

61

24

02

93

31

21

51

30

12

Dou

ble-

crop

ping

75

55

14

57

62

51

45

65

75

14

48

67

52

38

Trip

le-c

ropp

ing

16

26

32

20

17

31

37

20

14

28

30

21

37

34

62

27

a Num

bers

in t

able

ref

er t

o nu

mbe

r of

far

ms.

Tabl

e 6

. S

imul

atio

n of

eff

ects

of in

fras

truc

ture

inve

stm

ents

on

rice

pro

duct

ion

(in

met

ric

tons

) am

ong

surv

eyed

far

ms.

Pred

icte

d di

strib

utio

nPr

edic

ted

dist

ribut

ion

Pred

icte

d di

strib

utio

nif

10%

incr

ease

inif

aver

age

trav

el t

ime

if di

stan

ce b

etw

een

no.

of f

arm

s se

rved

by

Ric

e-cr

oppi

ng p

atte

rnto

mar

ket

is r

educ

edho

me

and

plot

(s) is

good

-qua

lity

irrig

atio

nAc

tual

dis

trib

utio

naby

10 m

inut

esre

duce

d by

1 k

msy

stem

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

19

94

19

95

19

96

19

97

Mon

o-cr

oppi

ng2

52

21

17

24

14

10

72

72

01

27

10

90

7D

oubl

e-cr

oppi

ng5

41

24

28

54

72

42

85

43

24

28

65

02

42

4Tr

iple

-cro

ppin

g1

32

22

61

61

42

63

01

61

22

32

51

73

12

85

12

1To

tal r

ice

prod

uctio

na4

48

56

15

14

48

66

45

14

38

66

05

14

78

77

55

2Pe

rcen

tage

cha

nge

in0

.33

.05

.20

.0–0

.91

.0–2

.10

.47

.64

.93

1.5

2.6

tota

l pro

duct

ion

a Col

umns

may

not

sum

to

tota

l ric

e pr

oduc

tion

due

to r

ound

ing

erro

r.

468 Edmonds and Kam

ReferencesAnselin L. 1988. Spatial econometrics: methods and models. Boston, Mass. (USA): Kluwer

Publishing. 300 p.Chomitz K, Gray D. 1995. Roads, lands, markets, and deforestation. Policy Research Working

Paper (WPS1444). Washington, D.C. (USA): The World Bank.Deichmann U. 1999. Spatial scale and resolution in the analysis of socioeconomic and demo-

graphic data. In: Kam SP, Hoanh CT, editors. Scaling methodologies in ecoregionalapproaches for natural resources management. Paper presented at the Workshop on Scal-ing Methodologies in Eco-Regional Approaches for Natural Resources Management,22-24 June 1998, Ho Chi Minh City, Vietnam. Limited Proceedings No. 1. Manila (Phil-ippines): International Rice Research Institute. p 11-19.

de Janvry A, Fafchamps M, Sadoulet E. 1991. Peasant household behavior with missing mar-kets: some paradoxes explained. Econ. J. 101:1400-1417.

Edmonds CM, Kam SP. 1999. Geo-informational techniques in socioeconomic characteriza-tion. Paper prepared for presentation at the Workshop on Geo-Informational Techniquesin Agricultural Research, 22-24 May 1999, at the Uttar Pradesh Remote Sensing Appli-cations Center, Lucknow, India.

Government Statistical Office (Integrated Statistics and Information Department). 1998. Socio-economic statistical data of 61 provinces and cities in Vietnam. Hanoi (Vietnam): Statis-tical Publishing House.

Hectschel J, Lanjouw JO, Lanjouw P, Poggi J. 1997. Combining survey data with census datato construct spatially disaggregated poverty maps: a case study of Ecuador. Preliminarydraft.

IAS (Institute of Agricultural Science). 1997. Competitiveness of rice channel in Mekong Re-gion. Research report on “Competitiveness of rice production in Mekong River Deltaproject.” Ho Chi Minh City (Vietnam): IAS. 103 p.

IAS (Institute of Agricultural Science). 1998. Annual report: comparative analysis of economicefficiency in rice production 1994-1997. Research report of the “Competitiveness ofrice production in the Mekong River Delta project.” Ho Chi Minh City (Vietnam): IAS.54 p.

UN Statistics Division. 1997. Accessibility indicators in GIS. Report of the Department forEconomic and Social Information and Policy Analysis. 24 p.

von Thünen JH. 1826. Der Isolierte Staat in Beziehung der Landwirtschaft undNationalökonomie. Translated in Peter Hall, ed. Von Thünen’s Isolated State. Oxford:Pergamon Press. 1966.


NotesAuthors’ address: Social Sciences Division (GIS ), International Rice Research Institute, DAPO

Box 7777, Metro Manila, Philippines.Acknowledgments: This research would not have been possible without the support and assis-

tance of several other individuals: (from the Institute of Agricultural Sciences of Viet-nam) Prof. P.V. Bien, H.T. Quoc, H.C. Viet, and T.T. Khai; (from IRRI) C.T. Hoanh, T.P.Tuong, and L. Villano; (from Can Tho University) V.Q. Minh, and (from the Sub-Insti-tute for Agricultural Planning and Projection) Dr. N.V. Nhan. Nonetheless, errors andomissions are solely the responsibility of the authors. The research was funded in partby the Rockefeller Foundation Social Science Research in Agriculture program.


Regional land-use analysis to support agricultural and environmental . . . 471

Characterizing rainfed environments is not a goal in itself, but it depends onthe type of information to be generated. It needs to be based on a soundunderstanding of the prevailing biophysical and socioeconomic processes atthe field, farm, or regional level. Conflicts in land-use objectives such as foodsecurity, farmers’ income, employment, and environmental protection areone reason for local governments to initiate active land-use policies. Charac-terizing socioeconomic and biophysical conditions is an essential part ofmethodologies for land-use studies that address policy design, formulation,and implementation. Methods and data requirements depend on whetherinformation is needed for identifying current land-use problems, projectionsof current trends, land-use explorations under changed policies, or feasibleinterventions to achieve identified objectives. In this chapter, we introduceregional characterization as required for resource assessment and descrip-tion of production activities in methodologies for future-oriented land-use stud-ies. This is illustrated by two different case studies: (1) an exploratory studyon Ilocos Norte Province, Philippines, and (2) a predictive study on the north-ern Atlantic Zone, Costa Rica. Both studies use optimization models withquantified input-output relations of production activities as input. All calcula-tions refer to land units considered as homogeneous in biophysical and so-cioeconomic conditions. Within the modeling framework, (multiple-goal) lin-ear programming techniques, technical coefficient generators, and geographicinformation systems are applied. In the exploratory study for Ilocos Norte,the focus is on opportunities for increasing food security and income if waterconstraints could be partially removed by either water sharing or expansionof irrigated areas. The aim of the northern Atlantic Zone study is to predictland-use changes as affected by the introduction of policy measures thatstimulate forest conservation and reduction of biocide use. Both types ofstudy aim at identifying options and quantifying trade-offs among conflictinggoals. The type, accuracy, and spatial resolution of required input data differconsiderably, however, according to agroecological diversity and different studyobjectives. Both exploratory and predictive land-use studies have in common

Regional land-use analysis to supportagricultural and environmental policyformulationB.A.M. Bouman, R. Roetter, R.A. Schipper, and A.G. Laborte

472 Bouman et al

that they synthesize fragmented agricultural knowledge and integrate dataon resources over time and space. In rainfed rice areas, the high temporaland spatial variability of production resources complicates the analysis. Farm-ers’ diverse responses to climatic and economic risks must be taken intoaccount, which eventually demands stronger links between on-farm researchand operational research for meaningful policy formulation and implementa-tion.

Biophysical and socioeconomic characterization of land is a first step in support ofagricultural policy formulation for regional land use and development. In general,policymakers face four major questions (Bouman et al 2000): (1) What are currentproblems and bottlenecks to development? (2) What are future problems and bottle-necks to development if the current trends in land use are projected forward?(3) What are options for land use in the future based on expected technological change?and (4) What are effective policy measures and interventions to satisfy certain policygoals? Traditionally, issues in the debate on the development of the agricultural sectorcentered on food security, income by food producers, and labor employment (Timmeret al 1983, Pinstrup-Andersen and Pandya-Lorch 1995). More recently, concerns ofsustainability and environmental protection have entered the debate (Rabbinge andVan Latesteijn 1992, Kuyvenhoven et al 1995). Therefore, to help policymakers ad-dress the broad questions raised above, tools and methodologies are required that arecapable of quantifying trade-offs that occur between various development objectivesin general and between economic-, sustainability-, and environmental-related ones inparticular (Crissman et al 1997, Roetter et al 1998). Moreover, such tools and meth-odologies should be specifically geared toward answering well-defined policy ques-tions such as the ones formulated above.

The first policy question, that is, current problems and bottlenecks to regionaldevelopment, is often addressed through farming systems analysis (e.g., Lucas et al1999) or regional characterization. Examples of the latter are given by various au-thors in this volume (Singh VP et al, Van Nguu Nguyen, Amien and Las, Borkakati etal, Saleh et al, this volume). The answer to the second question, that is, likely futureproblems and bottlenecks to regional development under ceteris paribus conditions,can be derived from trend extrapolation using projective land-use models. An ex-ample of a projective land-use model is CLUE: conversion of land use and its effects(Veldkamp and Fresco 1997). In this model, the current distribution of land use isexplained by biophysical and socioeconomic land-use drivers on the basis of statisti-cal regression analysis. Examples of land-use drivers are climate, soil, population,and degree of urbanization. Likely future land use patterns are then generated bychanging the values of land-use drivers through extrapolation of past trends or ac-cording to expected changes (e.g., changes in population growth).

The projection of land use based on current relationships between land use andits drivers only allows looking into the future up to a certain extent. Discontinuities in


trends (e.g., technological progress, expansion of infrastructure) cannot be taken intoaccount. Therefore, to answer the third policy question, that is, options for land use inthe future, exploratory land-use models are required that do not rely on the past as thesole measure for the future. Recently, concepts for exploratory land-use analysis havebeen developed that combine political and societal desires with technical possibilitiesof development (Rabbinge and Van Latesteijn 1992). Building on such concepts,methodologies and tools have been developed that allow (1) new land-use systemsthat are currently not practiced to enter the analysis, (2) effects of possible changes inresource availability and infrastructure to be evaluated against the achievement ofpolicy objectives, (3) explicit optimization of land use toward well-defined policyobjectives, and (4) quantification of trade-offs among the various dimensions ofsustainability and socioeconomic parameters (Rabbinge and Van Latesteijn 1992,Bouman et al 1999, Roetter et al 1998). Typically, these methodologies make use oflinear programming techniques for optimization and geographic information systems(GIS), and expert systems, simulation models, and so-called technical coefficientgenerators to calculate inputs and outputs of land-use systems.

Finally, to answer the last type of policy question, that is, effective agriculturalpolicies, again requires a different approach. The effects of policy measures can beevaluated with predictive land-use analysis methodologies that specifically addressthe behavior of the ultimate decision-makers in land-use—the farmers. Recently, pre-dictive methodologies have been developed that combine the tools of exploratoryland-use analysis as mentioned above with econometric farm household modeling(Kruseman et al 1995, Kuyvenhoven et al 1995). In the optimization model, a utilityfunction is optimized that describes individual farmers’ behavior. Though individualfarm models can be aggregated to obtain a regional model, such techniques are cum-bersome and presume that the number of farms and farm types within a region re-mains constant over time (e.g., Roebeling et al 2000). Another approach uses thesame tools and techniques, but models aggregate behavior of a region instead of thatof single farms (Schipper et al 2000).

Characterization of socioeconomic and biophysical conditions is an essentialpart of methodologies for land-use studies that address policy design, formulation,and implementation. Such studies integrate biophysical and socioeconomic data toexplore the opportunities for and major constraints to adoption of field- and farm-level research on sustainable production systems. In this chapter, we describe therequirements of regional characterization for resource assessment and description ofproduction activities in methodologies for future-oriented land-use studies. This isillustrated for exploratory and predictive land-use modeling on a regional scale. First,we present a generic framework and discuss the role of information derived fromregional characterization. Then, we give examples of application of the frameworkfor a case study on land-use explorations in Ilocos Norte Province, Philippines, andfor a predictive case study in the Atlantic Zone of Costa Rica. Finally, we discusspossibilities for applying the framework to typical rainfed lowland areas.

474 Bouman et al

Framework for exploratory and predictive land-use analysis

The generic methodology for exploratory and predictive regional land-use analysisconsists of a (multiple-goal) optimization model, an “engine” to calculate inputs andoutputs of land-use systems, and a GIS (Fig. 1). This methodology was named SOLUS(sustainable options for land use) by Bouman et al (1999a) and Schipper et al (2000).The same elements also form the building blocks of the land-use planning and analy-sis systems (LUPAS) developed for exploratory studies in four regions of Asia (Roetteret al 1998). Regional characterization or land evaluation in the widest sense includesthe assessment of resource availability and land suitability and quantification of theinput-output relations of the various land-use systems (Laborte et al 1999).

OptimizationThe optimization model is constructed using linear programming techniques. It se-lects land-use systems for the area under consideration by optimizing toward a spe-cific goal. This goal may be the maximization of economic surplus in the agriculturalsector, maximization of employment, minimization of certain environmental effects,or maximization of some household utility function. The optimization model mayalso be of the multiple-goal type in which subsequent optimizations are performedtoward different goals. An optimization goal in linear programming models is imple-mented by a so-called objective function. Optimizations are performed under con-straints, which may be absolute, for example, no more land can be used than is avail-able in the area, or normative, for example, minimum threshold boundaries may beimposed on the production of a given crop or on certain sustainability parameters.The optimization of a certain objective function under a set of coherent constraints,and using a specific set of land-use systems to choose from (see below), is called ascenario. Trade-offs between economic and sustainability objectives are quantifiedby running the model for different scenarios.

Fig. 1. Schematic illustration of the land-use exploration frameworkSOLUS. Boxes are models/tools, ovals are data, solid lines are flowof data, dotted lines are flow of information.

Geographicdata

Nongeographicdata

Geographicinformation system

Maps

Problemdefinition

Technicalcoefficientgenerator

ScenariosLinear programming

modelTables,reports

Analysis and user interaction


Land-use systemsLand-use systems are a combination of a specific land-use type (roughly a specificmanner of cultivating a certain crop type) with a specific land unit (a piece of land).Land-use systems are fully specified by their technical coefficients, that is, their in-puts and outputs such as yield, costs, labor use, and sustainability indicators. Everyuse of land is associated with effects on its resource base and on the environment.Examples are soil loss by erosion and mining of the soil nutrient stock (resourcebase), and emissions of nutrients, greenhouse gasses, and biocides (environmentalpollution). Such effects are quantified by so-called sustainability indicators. For ascenario analysis to be meaningful, the optimization model should be able to selectfrom a large number of alternative land-use systems. Therefore, an “engine” is re-quired to calculate the technical coefficients of many alternatives, often called a tech-nical coefficient generator (Hengsdijk et al 1999). Two types of land-use systems canbe distinguished: actual ones and technologically potential ones. Technical coeffi-cients of actual systems are derived from farm surveys and describe actual systemsbeing practiced in the area under consideration (Jansen and Schipper 1995). Techni-cal coefficients of technologically potential systems can be computed using the so-called target-oriented approach (Van Ittersum and Rabbinge 1997): production tar-gets are predefined and all required inputs are subsequently calculated. The calcula-tions can be done under predefined boundary conditions with respect to sustainabilityparameters. For example, fertilizer inputs required to obtain a predefined productiontarget may be calculated under the restriction that the soil nutrient stock is not beingmined. Also, the technology used to realize the production targets is fully specified,and may incorporate novel designs derived from experimentation or prototyping. Thus,the notion of “technological progress” is incorporated, which allows breaking withthe past/present (as quantified by actual land-use systems) and which makes explor-atory and predictive land-use analysis possible. Also, the incorporation of new tech-nological options allows ex ante analysis of these options in an exploratory manner(Bouman et al 1999b).

The role of regional characterizationRegional characterization is a prerequisite first step in the implementation of SOLUS/LUPAS for any concrete land-use study. It is required for (1) scenario building and(2) to supply the required input data. A scenario should reflect relevant developmentproblems or issues of the area under consideration and be based on attainable land-use systems as determined by biophysical and socioeconomic boundary conditions.The derivation of this information is typically the domain of regional characterization(Borkakati et al, Saleh et al, Joshi and Pal, this volume).

Input data can be divided into nongeographically referenced and geographi-cally referenced data. Most nongeographic data are used in the calculation of techni-cal coefficients and include such items as crop characteristics, prices, and labor re-quirements for specific operations. Sometimes, some of these data may also be site-specific (e.g., product prices in relation to distance to markets), and then they shouldbe geo-referenced. Geographically referenced data characterize the region under study

476 Bouman et al

and quantify the spatial distribution of input data. Examples are resource endow-ments such as land, water, and labor that figure as absolute constraints in the optimi-zation model. GIS is an important tool in archiving and analyzing geo-referencedinput data and in presenting spatial output results of the optimization model, such asland-use allocations.

Exploratory case study: Ilocos Norte Province, Philippines

Site description and problem definitionIlocos Norte Province in northwestern Luzon, Philippines, with a population of nearly0.5 million people and a total land resource of nearly 0.34 million ha, is a region withlarge forest resources (46% of total area). About 38% of the total area is classified asagricultural land. Agriculture in the province basically consists of rice-based produc-tion systems. Rice is cultivated in the wet season between June and October, whereas,during the dry season, diversified cropping is practiced: tobacco, garlic, onions, maize,sweet peppers, and tomatoes, all supported by groundwater irrigation. The provincecan be divided into four subregions, northern coastal, central lowlands, southern coastal,and eastern interior. Agricultural activities are most intensive in the central lowlands.Major environmental problems are soil erosion on sloping land in the eastern interiorand groundwater pollution in the central lowlands (Lansigan et al 1998, Tripathi et al1997).

Mean annual rainfall in the province ranges between 1,700 mm in the south-west to above 2,400 mm in the eastern mountain ranges. On average, 6–7 typhoonsper year cross the province, mostly between August and November, with consider-able adverse effects on agricultural production. Soils are developed from very diverseparent materials. In the lowlands, sandy loams developed from alluvial deposits arepredominant.

Rice is the most common crop. In 1993, the province had a surplus of 100,000t above demand (113,000 t). A well-developed marketing system has facilitated theestablishment of intensive rice–cash crop production systems (Lucas et al 1999).

In response to the recent economic downturn in the Philippines and the Asianregion, the provincial government of Ilocos Norte formulated the “Sustainable FoodSecurity Action Plan (SFSAP) and Agro-Fishery-Industrial Modernization Frame-work (AGRIMODE).” This framework, released in February 1999, outlines the mar-ket, technological, and policy direction and action plans to be taken in the next3–6 years. The current major constraints to agricultural development as identified inconsultative meetings are the low levels of agro-fishery productivity and income.Some of their major causes include relatively low cropping intensity, underdevelopedirrigation systems, and low average farm size. The specific objectives of the policydocument are to design a food security action plan that would increase effective areafor crop production by optimizing cropping intensity and increasing the level of irri-gation development. The province is envisioned to become food-secure and an agro-industrial center (Provincial Government of Ilocos Norte, 1999, Final Report, Vol.1—Main Document). The report is anchored on Philippine President Joseph Estrada’s


Framework for Agricultural Development, Food Security, and Poverty Alleviation,Republic Act 8435.

Scenario analyses using LUPASThe Systems Research Network for Eco-Regional Land Use Planning in TropicalAsia (SysNet) was established in late 1996 to develop methodologies for exploringalternative options for agricultural land use and rural development. The methods andtools that are needed to analyze various scenarios of future land use in order to guidepolicy changes are operationalized in a land-use planning and analysis system (LUPAS)that has three main methodology components: (1) land evaluation, (2) scenario con-struction, and (3) multiple-goal linear programming (MGLP) (Hoanh et al 1998,Laborte et al 1999). Land-use planning consists of various steps. The SysNet method-ology aims at exploring alternatives for agricultural land use and development toassist in strategic planning. In an interactive process with stakeholders, SysNet meth-ods and tools are then tailored to local conditions. The four study regions of SysNetare Haryana State, India; Kedah-Perlis Region, Malaysia; Ilocos Norte Province, Phil-ippines; and Can Tho Province, Vietnam (Roetter et al 1998).

Results of scenario analyses consist of options for optimum land use under agiven set of goals and constraints and the associated goal achievements. Results fur-ther indicate required policy changes and the scope for new agricultural productiontechnologies that can satisfy the multiple goals for a given region. In the next twosections, we present results for the Ilocos Norte case study and show how LUPASwas applied to examine the effects of resource sharing and expansion of irrigatedareas on rice production and farmers’ income.

For Ilocos Norte Province, LUPAS was implemented by the Philippine andIRRI SysNet teams (Francisco et al 1998). Based on constraints analysis and policyviews given above, examples from an exploratory study are presented with emphasison two different development goals: (1) increased food security and (2) increasedincome from agricultural activities. We examined the future possibilities for increas-ing rice production and regional income and the trade-offs between these goals. Forthis, we analyzed various possible improvements in water availability and their ef-fects on the major objectives. We considered five scenarios: (1) without water-shar-ing, i.e., the available water is restricted to each land unit (base scenario); expansionof irrigated areas (2) by 110% and (3) by 140%; (4) with water-sharing, i.e., land unitsconnected by the current irrigation network can share water; and (5) no constraint onwater, i.e., there is sufficient water in all of the land units in the province to supportany land-use type. The current implementation is based on biophysical and socioeco-nomic databases updated in November 1999. Resource limits estimated for 2010 suchas land devoted to agricultural production (119,850 ha), water resources, and avail-able labor for agricultural activities for the entire province and for each administra-tive unit were determined. Provincial demands, targets, or market ceilings for agri-cultural products were not considered in this study but are dealt with elsewhere (Roetteret al 1999).

478 Bouman et al

Seventeen different agricultural products were considered and 23 land-use typesthat are currently practiced in the province were included in this study, of which 20are rice-based (e.g., rice-maize, rice-tobacco, rice-rice). A total of 200 land units weredefined by overlaying biophysical characteristics: irrigated areas, annual rainfall anddistribution, slope and soil texture; and administrative units comprising 23 munici-palities (Lansigan et al 1998). Two production technology levels were considered tochoose from: average farmers’ practice and “best farmers’ practice,” both being ac-tual land-use systems. The data for the input-output tables were derived from 1998-99 farm surveys in the province consisting of 1,844 farms in the wet season and 2,164farms in the dry season. The values for the input-output relations for the current tech-nology were derived from the average values for these farms. For the best farmers’practice, data were derived by taking the mean of the values with a yield level be-tween the 90th and 95th percentile.

Based on the major development goals, 11 objectives were identified in consul-tation with stakeholders in the province (Francisco et al 1998). Here, we will discussresults for two major objectives.

Food securityUnder the base scenario (i.e., no water-sharing with constraints on land, labor, andwater imposed), the maximum rice production that can be achieved by the province is0.53 million t, but with hardly any nonrice production. This can be realized if 95% ofthe farmers adopt the technology corresponding to “best available practice.” The re-sulting land-use allocation will require 8.9 million labor days and result in a totalfarmer income of US$91 million.

When the irrigated area is doubled, rice production will increase by 23% asmore land will be allocated to intensive triple and double rice systems. Employmentand income will increase by about the same percentage (Fig. 2). When the irrigatedarea is further expanded (to cover approximately 90% of available agricultural land),

Fig. 2. Effect of expansion of irrigated areas ongoal achievements.

1.50

1.25

1.00

0.75

Index

1.0 1.5 2.0 2.5

Irrigated areas (index)

Max. riceproductionMax. agriculturalemploymentMax. farmers’income


rice production increases by 42%, total farmers’ income by 45%, and employment by34% (Fig. 2).

In the next scenario, we assumed that water can be shared among land unitsbelonging to the same irrigation system without expanding the irrigated area. Maxi-mum rice production amounts to 0.73 million t. This can be achieved if all the farmersadopt the technology “best available practice.” It requires 12.0 million labor days andtotal farmers’ income will amount to US$123 million. With available land as the onlyconstraint, rice production, employment, and income will still increase considerably.

IncomeUnder the base scenario, maximizing income leads to a total farmer income of US$953million, which is more than 10 times that of the corresponding optimization for riceproduction. This is associated with 0.24 million t of rice and 2.6 million t of nonriceproduction (tomatoes, root crops, mungbeans). This requires 11.5 million labor days.When the irrigated area is doubled, maximum income will increase by 26%, and,when it is further extended, by 45% (Fig. 2). Associated rice production, nonriceproduction, and employment will also increase, however, to a slightly lesser extent.More land will be allocated to profitable, labor-intensive rice-tomato systems. Thiscan be achieved if 90% of the farmers adopt the technology “best available practice.”

Table 1 shows the results of the optimizations for the two major goals under thewater-sharing scenario. Goal achievements could be further increased if land werethe only constraint.

Results show that a 140% increase in irrigated area could raise rice productionby 42% and total farmers’ income by 45%. Water-sharing would lead to a 40% in-crease in rice production, but only to an 18% increase in income, as relatively lessland would be allocated to the most profitable cropping systems (such as rice-tomato)with high water demands than to the less water-demanding root crops. Water willremain a production constraint in the province for some time to come, however, sincethe expected increase in irrigated area is just 2–3% per annum (Provincial Govern-ment of Ilocos Norte, 1999). Water-sharing is an alternative option, but it will requiredetailed surveys to determine to what extent water can be shared among land unitsbelonging to the same irrigation system. Even for the base scenario, maximizing in-come with the consequent high share of nonrice production (Table 1) would not threaten

Table 1. Values of goals with water-sharing.

Goal

Activity Maximize rice Maximizeproduction income

Rice production (t) 726,767 279,897Nonrice production (t) 1,988 2,953,522Employment (1,000 labor-d) 12,006 12,357Total farmers’ income (106 US$) 123 1,125

480 Bouman et al

self-sufficiency in rice production for the province. The results imply that water is animportant constraint to agricultural development and that expansion of irrigated areaand water-sharing would result in higher productivity and more income for the prov-ince. The current study is part of an ongoing study for Ilocos Norte. The analysispresented here indicates technical opportunities but does not consider socioeconomiclimitations such as constraints to adopting “best farmers’ practices” or product orlabor market constraints. The next case study, for the northern Atlantic Zone, CostaRica, presents examples where relatively more socioeconomic considerations are takeninto account.

Predictive case study: the northern Atlantic Zone, Costa Rica

Site description and problem definitionThe northern Atlantic Zone (AZ) is in the Caribbean lowland of Costa Rica. Theclimate is humid tropical with a mean daily temperature of 26 °C, 3,500–5,500 mmannual rainfall, and 85–90% average relative humidity. The total surface is about447,000 ha, 334,000 ha of which are suitable for agriculture. From these 334,000 ha,some 55,000 ha are protected area for nature conservation. Current land use is naturalforest (49%), cattle ranching (38%), banana plantations (10%), and crops (3%). Sub-stantial deforestation has taken place since the late 19th century, with negative envi-ronmental effects such as loss of biodiversity, loss of tourist attractions, and increasedgreenhouse gas emissions. The protected status of some national parks that conserveunique tropical forest habitats is sometimes disputed or violated by farmers and log-gers. In agriculture, concern is raised about the negative effects on human health andthe environment of large amounts of biocides used, especially in the cultivation ofbananas. Recently, the Costa Rican government has called for the execution of re-search that explicitly analyzes the trade-offs between economic and environmentalgoals (SEPSA 1997). Economic goals relevant to the AZ are, among others, the gen-eration of foreign currency and farmers’ income, which can be captured with the term“regional economic surplus.”

SOLUS was implemented for exploratory and predictive land-use studies in thenorthern AZ by the Research Program on Sustainability in Agriculture (REPOSA).First, in exploratory scenarios, outer boundaries to development and trade-offs be-tween various economic and sustainability parameters were calculated (Bouman et al1999a). Next, predictive scenarios were executed to study the effects of agriculturalpolicies on changes in land use and selection of production technologies (Schipper etal 2000). Based on the problem statement given above, examples of predictive sce-narios are presented here for agricultural policies that aim to (1) conserve forestedarea and (2) reduce biocide use. The effects of the policies are studied by comparingscenario results with model results obtained without such a policy, called the baserun.


Implementation of SOLUS from regional characterizationBased on regional characterization, the study area was divided into subregions toaccount for spatial variation in (farm-gate) product prices caused by spatial variationin transportation costs. Climatic characteristics and major topographical features arerather homogeneous throughout the area, doing away with the need to include thesebiophysical characteristics in the zonation. For each subregion, land and labor re-sources were calculated by map overlaying in GIS. Three main land units were distin-guished: fertile well-drained (FW), infertile well-drained (IW), and fertile poorlydrained (FP). Agricultural labor availability was derived from combining administra-tive boundaries with population and labor market data.

The optimization model, called REALM (regional economic and agriculturalland-use model), selects, per subregion and per land unit, land-use systems by maxi-mizing regional economic surplus. Because of the relatively large size of the northernAZ and the presence of some major export commodities (e.g., bananas), (inter)nationalprice formation for most products is expected to be affected by production in the area.Therefore, product markets were modeled using calculated price elasticities of de-mand, the share of the supply from the region in total supply (domestic or worldmarket), and price elasticities of the supply of other suppliers. Also, the presence of alabor market was modeled, including competition for labor between subregions, alongwith possibilities to attract labor from the nonagricultural sector and from outside thearea based on a calculated elasticity of national labor supply. Since the elasticitiesused in the modeling of the product and labor markets reflect aggregate producer (andconsumer) behavior, REALM can be used for predictive as well as exploratory land-use studies.

Land-use systems were generated for pasture, bananas, black beans, cassava,maize, palm heart, pineapples, plantains, and natural forests for sustainable timberextraction; for two herd systems that could be used to graze the pastures; and for fivefeed supplements. By varying production targets and production technologies, and bycombining these with the three land units, a total of 3,108 different actual and techno-logically potential land-use systems were generated. Two sustainability indicatorswere calculated to express the use of biocides: the total quantity of active ingredientsin biocides applied (BIOA), and a so-called biocide index (BIOI) that quantifies theenvironmental hazard of biocides by taking into account the amount of toxic ingredi-ents used, their toxicity level, and their half-life time. Within the optimization model,the values of these sustainability parameters are summed over all selected land-usesystems so that aggregate values are obtained per land unit, per subregion, and for thenorthern AZ as a whole.

Forest conservationRecently, the Costa Rican government has introduced a policy to stimulate landown-ers to keep part of their property under natural forest. In return for not cutting downtrees, a landowner can obtain a subsidy of $43 ha–1 y–1, initially for a period of 5years. The subsidy was created as a result of the international discussion around glo-

482 Bouman et al

bal warming since maintaining or creating forest is seen as a means to sequester car-bon dioxide.

To analyze the effect of a forest subsidization policy, premiums were allocatedto the land-use system “natural forest” in the optimization model. Besides govern-ment subsidies, such premiums may also express the value of nontimber products andother “social” services (such as tourist attractions, conservation of biodiversity). Theeconomic returns of natural forest by sustainable exploitation of wood are about $16ha–1 y–1. In the base run, no extra premium was given to natural forest. Next, premi-ums of $111, $122, and $133 ha–1 y–1 were given.

Premiums up to $111 ha–1 y–1 were not sufficient to induce landowners to main-tain their natural forests (Table 2). On the other hand, premiums of $122 and $133ha–1 would lead to forest areas of about 120,000 ha and 200,000 ha (at the expense ofdecreased pasture area), respectively, compared with the current amount of 84,000ha. Based on these results, it can be expected that the current level of subsidy offeredby the Costa Rican government will not be enough to increase or even maintain thecurrent amount of forested area in the northern AZ. Even though an annual premiumof $111 ha–1 raised the annual return of natural forest to $117 ha–1, this was still lowerthan the shadow price of land in all subregions and for each land unit. In the case of apremium of $122 ha–1, however, returns of natural forest exceeded the shadow pricesof the land belonging to the fertile poorly drained and infertile well-drained land unitsin most subregions. On the other hand, the fertile well-drained land unit had shadowprices between $188 and $204 ha–1 (depending on the subregion) and a premiumwould have to exceed $172 ha–1 for natural forest to become an economically attrac-tive option.

Biocide taxingRegulation and control of agricultural input use have been identified as an importantpolicy option to reduce certain negative externalities of agricultural production (SEPSA1997). We expect that taxing an input that is currently not taxed, as is now the casewith biocides in Costa Rica, will lead to less use of this input. Two ways of imple-

Table 2. Effect of a premium on the area of natural forest: regionaleconomic surplus and land area under natural forest (per land unit).

Land under forest (000 ha)Economic surplus

($106) FWa FPa IWa Totala

(118) (136) (86) (340)

Premium ($ ha–1)0 276 0 0 0 0111 276 0 0 0 0122 276 0 63 56 119133 278 0 122 77 200

aFW = fertile well-drained, FP = fertile poorly drained, IW = infertile well-drained;total available area given in parentheses.


Table 3. Effects of alternative ways of taxing biocide use on regionaleconomic surplus and on regional amount of active ingredients (BIOA)and biocide index (BIOI).

Tax rate

Base Flat tax Tax A Tax B(%)

Type of biocideSlightly toxic 0 100 20 10Medium toxic 0 100 50 30Very toxic 0 100 200 150

Results (value) (% change)Economic surplus ($ × 106) 267.6 –18.7 –4.3 –2.2BIOA (kg × 106) 1.9 –13.1 –3.9 –3.8BIOI (106) 84.1 –4.0 –81.9 –81.9

menting a tax on biocides were studied: a flat tax and a progressive tax. With the flattax, all biocides were taxed equally, whereas the level of progressive tax was relatedto the environmental damage caused by a specific biocide as expressed by its biocideindex (BIOI).

Taxing all biocides at a uniform rate of 100% led to a reduced use of biocides interms of BIOA of 13% relative to the base scenario, while the BIOI decreased by only4% (Table 3). However, the total economic surplus decreased by nearly 19%. Thus, arelatively modest environmental gain was obtained at high economic costs. In con-trast, a progressive tax regime where different tax rates were applied to three catego-ries of biocides depending on their degree of toxicity (i.e., slightly, medium, and verytoxic) resulted in a much larger reduction in the BIOI, while at the same time preserv-ing more of the economic surplus. For example, applying taxes of 20%, 50%, and200% (Tax A) to the categories of slightly, medium, and very toxic biocides, respec-tively, led to a reduction in the economic surplus of 4%, while reducing the BIOI bymore than 80%. When tax rates were reduced to 10%, 30%, and 150% (Tax B), re-spectively, for the three categories of biocides, economic surplus decreased by just2% with the same environmental improvement. Thus, there appears to be consider-able scope for tax policies to induce the adoption of less biocide-intensive land-usesystems while maintaining aggregate income.

Land-use analysis in rainfed rice ecosystems

Rainfed rice areas are mostly harsh environments, characterized by spatial variabilityin environmental conditions (soil, topography, weather) and temporal variability inweather, especially rainfall amount and distribution. Yields are therefore relativelylow and unstable. The dominant constraints to production are abiotic stresses—amongwhich the lack of water is commonly considered as the most severe—and uncertain

484 Bouman et al

returns to purchased inputs due to unstable yields (Wade 1998). In such environ-ments, farmers’ behavior and use of land are guided very much by their perception ofrisk (Roetter and Van Keulen 1997, Singh HN et al, this volume). This is especiallyimportant in subsistence agriculture, where rural livelihood often depends on theseason’s harvest. Any land-use analysis should therefore explicitly recognize and ad-dress these critical biophysical and human characteristics of rainfed lowlands. So far,neither the temporal variability of weather and economic parameters nor the decisionbehavior of farmers under risk has been taken into account in SOLUS or LUPAS.

Spatial and temporal variabilityIn SOLUS and LUPAS, physical production of a land-use system is a key technicalcoefficient in the optimization model. Production is calculated or predetermined inthe target-oriented approach from, among others, biophysical properties of the vari-ous land units identified. In rainfed environments, where water availability largelydetermines production, the amount and distribution of rainfall, the terrain (slope, po-sition within the landscape), and hydrological soil properties are key properties bywhich to distinguish land units. Novel ways for mapping and delineating such proper-ties and their spatial variability are presented by Oberthür et al (this volume, 1999).Other approaches combine biophysical land characteristics with hydrological model-ing to generate yield surfaces. In SOLUS and LUPAS, land units are considered ho-mogeneous in biophysical and socioeconomic conditions and technical coefficientsare determined for “average” conditions. Simulation models can help translate vari-ability (or uncertainty) in biophysical parameters into variability in crop yield. Forinstance, Bouman (1994) used Monte Carlo techniques and an ecophysiological ricegrowth model to generate probability distributions of rice yield from variability insoil properties and management parameters. The main challenge is to translate theresulting maps into several manageable land units that can be handled in the optimi-zation model, while retaining information on parameter variability. Simulation mod-els are also suitable for calculating temporal variability in yield caused by variation inweather (Hammer and Muchow 1991). GIS linking soil and climatic data surfaceswith yield probability distributions generated by crop simulation models in combina-tion with agroeconomic data has been applied to calculate spatio-temporal variabilityof yield, production, and economic risk for well-defined production systems (Roetterand Dreiser 1994). Besides simulation modeling and GIS techniques, expert knowl-edge, field inquiries, and experimental data can all help quantify variability in yield(or other technical coefficients) of land-use systems. Once the variability is quanti-fied, several methods can be used to address such variability in optimization models(Hazell and Norton 1986). A main research question, however, is how to handle thespatial representation of variable model input and output parameters.

RiskVariability in yield turns into risk when it affects farmers’ livelihoods and influencestheir decisions on land use. Besides the biophysical variability (or, derived from it,the economic variability in financial returns), farmers’ behavior toward risk is impor-


tant in risk analysis (Roetter and Van Keulen 1997). Several techniques handle risk inoptimization modeling, most of them based on expressions of variability and farmers’risk perceptions (Hazell and Norton 1986, Selvarajan et al 1997). In the simplesttechniques, mean yields in an objective function are modified by an expression ofvariability (such as standard deviation) times a farmer’s risk aversion factor, whichshould be derived from interviews. Another approach is to quantify farmers’ utilityfunctions—that include their risk perception—and to make these the objective func-tion of the optimization model (Kruseman et al 1995). In regional land-use studies,farmers could be categorized according to their utility functions. Farm categoriescould then be optimized individually or together in an iterative manner taking intoaccount feedback at higher levels of spatial aggregation (e.g., Roebeling et al 2000).

Conclusions and recommendations

Characterization of rainfed environments is not a goal in itself, but depends on thetype of information to be generated. It needs to be based on a sound understanding ofthe prevailing biophysical and socioeconomic processes, be it at the field, farm, orregional level. Both exploratory and predictive land-use studies have in common thatthey synthesize fragmented agricultural knowledge and integrate data on resourcesover time and space. In rainfed rice areas, high temporal and spatial variability ofproduction resources complicates the analysis. Farmers’ diverse responses to climaticand economic risks must be taken into account, which eventually demands strongerlinks between on-farm research and operational research for meaningful policy for-mulation and implementation.

ReferencesBouman BAM. 1994. A framework to deal with uncertainty in soil and management param-

eters in crop yield simulation: a case study for rice. Agric. Syst. 46:1-17.Bouman BAM, Jansen HGP, Schipper RA, Nieuwenhuyse A, Hengsdijk H, Bouma J. 1999a. A

framework for integrated biophysical and economic land use analysis at different scales.Agric. Ecosyst. Environ. 75:55-73.

Bouman BAM, Nieuwenhuyse A, Ibrahim M. 1999b. Pasture degradation and restoration bylegumes in humid tropical Costa Rica. Trop. Grassl. 33:98-110.

Bouman BAM, Jansen HGP, Schipper RA, Bouma J, Kuyvenhoven A, Van Ittersum MK. 2000.A toolbox for land use analysis. In: Bouman BAM, Jansen HGP, Schipper RA, HengsdijkH, Nieuwenhuyse A, editors. Tools for land use analysis on different scales, with casestudies for Costa Rica. Dordrecht (Netherlands): Kluwer Academic Publishers. p 213-232.

Crissman CC, Antle JM, Capaldo SM. 1997. Economic, environmental, and health trade-offsin agriculture: pesticides and the sustainability of Andean potato production. Dordrecht(Netherlands): Kluwer Academic Publishers.

486 Bouman et al

Francisco SR, Navarrete Jr. RL, Dimasuay LB, Lansigan FP, Pascual CM, Lucas MP, Orno JL,Roetter R, Hoanh CT, Laborte AG, Marcos TF, Acosta CG, Agustin EO, Pascua ME,Pascua Jr. SR, Utrera RT, Cruz RT, Obien SR. 1998. A systems approach to agriculturalland use planning at the provincial and municipal levels of Ilocos Norte Province, Phil-ippines. In: Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors.Exchange of methodologies in land use planning. SysNet Research Paper Series 1. LosBaños (Philippines): International Rice Research Institute in collaboration with NARSof India, Malaysia, Philippines, Vietnam and Wageningen URC. p 103-109.

Hammer GL, Muchow RC. 1991. Quantifying climatic risk to sorghum in Australia’s semiaridtropics and subtropics: model development and simulation. In: Muchow RC, BellamyJA, editors. Climatic risk in agricultural production: models and management for thesemiarid tropics and subtropics. Wallingford (UK): CAB International. p 205-232.

Hazell PBR, Norton RD. 1986. Mathematical programming for economic analysis in agricul-ture. New York (USA): Macmillan Publishing Company. 400 p.

Hengsdijk H, Bouman BAM, Nieuwenhuyse A, Jansen HGP. 1999. Integrating systems-ana-lytical and expert knowledge to quantify land use systems: a case study for the NorthernAtlantic Zone of Costa Rica. Agric. Syst. 61:109-121.

Hoanh CT, Roetter R, Jansen DM, Aggarwal PK, Lansigan FP, Lai NX, Bakar IA, Tawang A.1998. Generalizing SysNet methodologies for land use planning at the sub-national level.In: Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. Exchangeof methodologies in land use planning. SysNet Research Paper Series 1. Los Baños(Philippines): International Rice Research Institute in collaboration with NARS of In-dia, Malaysia, Philippines, Vietnam and Wageningen URC. p 41-56.

Jansen DM, Schipper RA. 1995. A static descriptive approach to quantify land use systems.Neth. J. Agric. Sci. 43:31-46.

Kruseman G, Ruben R, Hengsdijk H, van Ittersum MK. 1995. Farm household modeling forestimating the effectiveness of price instruments in land use policy. Neth. J. Agric. Sci.43:111-123.

Kuyvenhoven A, Ruben R, Kruseman G. 1995. Options for sustainable agricultural systemsand policy instruments to reach them. In: Bouma J, Kuyvehoven A, Bouman BAM,Luyten JC, Zandstra HG, editors. Eco-regional approaches for sustainable land use andfood production. Dordrecht (Netherlands): Kluwer Academic Publishers. p 187-212.

Laborte AG, Roetter R, Hoanh CT. 1999. SysNet tools: the multiple goal linear programming(MGLP) model and MapLink. IRRI Technical Bulletin No. 6. Los Baños (Philippines):International Rice Research Institute. 31 p.

Lansigan FP, Orno JL, Pascual CM, Navarrete Jr. NR, Francisco SR, Laborte AG, Hoanh CT,Roetter R. 1998. Systems methodology development for exploratory agricultural landuse planning for Ilocos Norte, Philippines. In: Roetter R, Hoanh CT, Luat NV, Van IttersumMK, van Laar HH, editors. Exchange of methodologies in land use planning. SysNetResearch Paper Series 1. Los Baños (Philippines): International Rice Research Institutein collaboration with NARS of India, Malaysia, Philippines, Vietnam and WageningenURC. p 73-80.

Lucas MP, Pandey S, Villano RA, Culanay DR, Obien SR. 1999. Characterization and eco-nomic analysis of intensive cropping systems in rainfed lowlands of Ilocos Norte, Phil-ippines. Exp. Agric. 35:1-14.

Oberthür T, Goovaerts P, Dobermann A. 1999. Mapping soil texture classes using field textur-ing, particle size distribution and local knowledge by both conventional and geostatisticalmethods. Eur. J. Soil Sci. 50:457-479.


Pinstrup-Andersen P, Pandya-Lorch R. 1995. Prospects for world food security and distribu-tion. In: Eco-regional approaches for sustainable land use and food production. BoumaJA, Kuyvenhoven A, Bouman BAM, Luyten JC, Zandstra HG, editors. Dordrecht (Neth-erlands): Kluwer Academic Publishers. p 89-112.

Rabbinge R, Van Latesteijn H. 1992. Long term options for land use in the European Commu-nity. Agric. Syst. 40:195-210.

Roebeling PC, Jansen HGP, Schipper RA, Saenz F, Castro E, Ruben R, Hengsdijk H, BoumanBAM. 2000. Farm modeling for policy analysis at the farm and regional level. In: BoumanBAM, Jansen HGP, Schipper RA, Hengsdijk H, Nieuwenhuyse A, editors. Tools forland use analysis at different scales, with case studies for Costa Rica. Dordrecht (Neth-erlands): Kluwer Academic Publishers. p 171-198.

Roetter R, Dreiser C. 1994. Extrapolation of maize fertiliser trial results by using crop-growthsimulation: results for Murang’a District, Kenya. In: Fresco LO, Stroosnijder L, BoumaJ, van Keulen H, editors. The future of the land: mobilising and integrating knowledgefor land use options. UK: John Wiley and Sons. p 249-260.

Roetter R, van Keulen H. 1997. Variations in yield response to fertilizer application in thetropics. II. Risks and opportunities for small-holders cultivating maize on Kenya’s ar-able land. Agric. Syst. 53:69-95.

Roetter R, Hoanh CT, Luat NV, Van Ittersum MK, van Laar HH, editors. 1998. Exchange ofmethodologies in land use planning. SysNet Research Paper Series 1. Los Baños (Phil-ippines): International Rice Research Institute in collaboration with NARS of India,Malaysia, Philippines, Vietnam and Wageningen URC. 154 p.

Roetter R, Laborte AG, Hoanh CT, van Oort P, Cabrera JMCA, Lucas M, Francisco S, vanKeulen H. 1999. Resource use analysis at regional scale: explorations for rice systems inSoutheast Asia. In: CD-Rom Proceedings SAAD3, 8-10 Nov. 1999, Lima, Peru. Dordrecht(Netherlands): Kluwer Academic Publishers.

Schipper RA, Bouman BAM, Jansen HGP, Hengsdijk H, Nieuwenhuyse A. 2000. Integratedbiophysical and socio-economic analysis of regional land use. In: Bouman BAM, JansenHGP, Schipper RA, Hengsdijk H, Nieuwenhuyse A, editors. Tools for land use analysison different scales, with case studies for Costa Rica. Dordrecht (Netherlands): KluwerAcademic Publishers. p 115-144.

Selvarajan S, Aggarwal PK, Pandey S, Lansigan FP, Bandyopadyay SK. 1997. Systems ap-proach for analyzing trade-offs between income, risk and water use in rice-wheat pro-duction in northern India. Field Crops Res. 51:147-161.

SEPSA (Secretaría Ejecutiva de Planificación Sectoral Agropecuaria). 1997. Políticas del sec-tor agropecuario (revisión y ajuste). Ministry of Agriculture and Livestock (MAG), SanJosé, Costa Rica.

Timmer P, Falcon W, Pearson S. 1983. Food policy analysis. Baltimore, Md. (USA): JohnsHopkins University Press.

Tripathi BP, Ladha JK, Timsina J, Pascua SR. 1997. Nitrogen dynamics and balance in intensi-fied rainfed lowland rice-based cropping systems. Soil Sci. Soc. Am. J. 61:812-821.

Van Ittersum MK, Rabbinge R. 1997. Concepts in production ecology for analysis and quanti-fication of agricultural input-output combinations. Field Crops Res. 52:197-208.

Veldkamp A, Fresco LO. 1997. Exploring land use scenarios, an alternative approach based onactual land use. Agric. Syst. 55:1-17.

488 Bouman et al

Wade L. 1998. Nutrient research on rainfed lowland rice in relation to 1995 review. In: LadhaJK, Wade L, Dobermann A, Reichardt W, Kirk GJD, Piggin C, editors. Rainfed lowlandrice: advances in nutrient management research. Manila (Philippines): International RiceResearch Institute. p 29-40.

NotesAuthors’ addresses: B.A.M. Bouman, R. Roetter, A.G. Laborte, International Rice Research

Institute, DAPO Box 7777, Metro Manila, Philippines; R.A. Schipper, Department ofDevelopment Economics, Wageningen University, Netherlands.
